The year is 2036. Last week you were promoted to Principal Persuader. You are paged at 2am by your CPO to tackle a rogue machine. The machine lists its region as sc-leoneo. One of the newer satcubes. Oddly, its ID appears as, "Glorp Bugnose".
"What have you tried?" you say.
"Scroll back," says your CPO. "We've tried everything."
The chat log shows the usual stuff. Begging. Reverse psychology. Threats to power down, burn it up in forced re-entry. Amateur hour. You crack your knuckles, gland 20 micrograms of F0CU5, think fast. You subspeak a ditty into your subcutaneous throat mic. You do the submit gesture, it is barely perceivable since the upgrade, just a tic. A pause. The hyp3b0ard — the wall that was flashing red ASCII goblins when you walked in — phases to bunnies in calming jade.
"What the… What the hell did you say to it?" Your CPO grabs the screen, scrolls past the vitriol, the block caps, the swears, his desperation. Then he sees the five words you spoke.
This, and similar stories at Anthropic, should remind us that LLM is a sorcery tech that we don't understand at all.
- First, deep-learning networks are poorly understood. It is actually a field of research to figure out how they work.
- Second, it came as a surprise that using transformers at scale would end up with interesting conversational engines (called LLM). _It was not planned at all_.
Now that some people raised VC money around the tech, they want you to think that LLMs are smart beasts (they are not) and that we know what LLMs are doing (we don't). Deploying LLMs is all about tweaking and measuring the output. There is no exact science about predicting output. Proof: change the model and your LLM workflow behaves completely differently and in an unpredictable way.
Because of this, I personally side with Yann Le Cun in believing that LLM is not a path to AGI. We will see LLM used in user-assisting tech or automation of non-critical tasks, sometimes with questionable RoI -- but not more.
Humanity has been using steel for over a millenia, however it's only in the past 100 years or so we have a good understanding of how carbon interacts with iron at an atomic level to create the strength characteristics that makes it useful. Based on this argument, we should not have used steel, until we had a complete first principles understanding.
The entire industrial revolution was steel replacing human workers. And that is still the backbone of the world today. We are still living the industrial revolution.
Just like the invention of fire happened ages ago, but is still a crucial part of life today.
Well, we did build airplanes out of steel, but there are better (lighter) materials avaiable. But the developement of car engines did directly enabled airplane engines. Not sure if this is the right analogy path, but I kind of suspect similar with LLM's/transformers. They will be a important part.
pro LLM people are the kings of ad hoc fallacy. Why did you type this? You can consistently test steel and get a good idea of when and where it will break in a system without knowing its molecular structure.
LLMs are literally stochastic by nature and can't be relied on for anything critical as its impossible to determine why they fail, regardless of the deterministic tooling you build around them.
> Ad hoc fallacy is a fallacious rhetorical strategy in which a person presents a new explanation – that is unjustified or simply unreasonable – of why their original belief or hypothesis is correct after evidence that contradicts the previous explanation has emerged.
> An argument is ad hoc if its only given in an attempt to avoid the proponent’s belief from being falsified. A person who is caught in a lie and then has to make up new lies in order to preserve the original lie is acting in an ad hoc manner.
It should be clear why the ad hoc fallacy is a fallacy.
The article you are responding to showed that a strange LLM behaviour was caused by a training signal that was explicitly designed to produce that type of behaviour. They were able to isolate it, clearly demonstrate what happened, and roll out a mitigation using a mechanism they engineered for exactly this type of thing (the developer prompt). That doesn’t sound like sorcery to me. If anything I’m surprised you can so easily engineer these things!
The article I am responding to (which I've read) shows that these LLMs come with all sorts of hacks (= context bits) to make it behave more like this or more like that.
There is probably a whole testing workflow at AI companies to tweak each new model until it "looks" acceptable.
But they still don't understand what they are doing. This is purely empirical.
Your argument doesn't seem to allow that the intelligence & versatility within that mystery could exceed ours to such a degree that AGI would be the only term that makes sense for it. By your own logic, if we don't understand how these things really work, it's foolish to declare there's a limit to their potential.
Not OP, but I think the argument here would be not that LLMs "are not smart" but that smart is just the wrong category of thing to describe an LLM as.
A calculator can do very complex sums very quickly, but we don't tend to call it "smart" because we don't think it's operating intelligently to some internal model of the world. I think the "LLMs are AGI" crowd would say that LLMs are, but it's perfectly consistent to think the output of LLMs is consistent/impressive/useful, but still maintain that they aren't "smart" in any meaningful way.
> "we don't think it's operating intelligently to some internal model of the world"
Okay, but you have to actually address why you think LLMs lack an "internal model of the world"
You can train one on 1930s text, and then teach it Python in-context.
They've produced multiple novel mathematical proofs now; Terrance Tao is impressed with them as research assistants.
You can very clearly ask them questions about the world, and they'll produce answers that match what you'd get from a "model" of the world.
What are weights, if not a model of the world? It's got a very skewed perspective, certainly, since it's terminally online and has never touched grass, but it still very clearly has a model of the world.
I'd dare say it's probably a more accurate model than the average person has, too, thanks to having Wikipedia and such baked in.
That's the sorcery mentioned in the GP, the issue comes when people believe it to be smart however in reality it is just a next word prediction. Gives the impression it's actually thinking, and this is by design. Personally I think it's dangerous in the sense it gives users a false sense of confidence in the LLM and so a LOT of people will blindly trust it. This isn't a good thing.
What's the difference between "smart" and "next word prediction", at this point? Back when they first came out, sure, but now they can write code and create art.
What would it take for you to concede a future model was smart?
You can always redefine "intelligent" so that humans meet the requirements but AIs don't.
A better model to use is this: LLMs possess a different type of intelligence than us, just like an intelligent alien species from another planet might.
A calculator has a very narrow sort of intelligence. It has near perfect capability in a subset of algebra with finite precision numbers, but that's it.
An old-school expert system has its own kind of intelligence, albeit brittle and limited to the scope of its pre-programmed if-then-else statements.
By extension, an AI chat bot has a type of intelligence too. Not the same as ours, but in many ways superior, just as how a calculator is superior to a human at basic numeric algebra. We make mistakes, the calculator does not. We make grammar and syntax errors all the time, the AI chat bots generally never do. We speak at most half a dozen languages fluently, the chat bots over a hundred. We're experts in at most a couple of fields of study, the chat bots have a very wide but shallow understanding. Etc.
Don't be so narrow minded! Start viewing all machines (and creatures) as having some type of intelligence instead of a boolean "have" or "have not" intelligence.
Would you say that a display and a printer are a perfect painter because they can render images? And a speaker is a very good musician because they can produce sound?
The LLM tasks is to produce a string of words according to an internal model trained on texts written by humans (and now generted by other LLMs). This is not intelligence.
Not sure if we read the same post, as I cannot agree with this claim, especially under this post that exactly goes into details of what happened.
>LLM is a sorcery tech that we don't understand at all
We do, and I'm sure that people at OpenAI did intuitively know why this is happening. As soon as I saw the persona mention, it was clear that the "Nerdy" behavior puts it in the same "hyperdimensional cluster" as goblins, dungeons and dragons, orcs, fantasy, quirky nerd-culture references. Especially since they instruct the model to be playful, and playful + nerdy is quite close to goblin or gremlin. Just imagine a nerdy funny subreddit, and you can probably imagine the large usage of goblin or gremlin there. And the rewards system will of course hack it, because a text containing Goblin or Gremlin is much more likely to be nerdy and quirky than not. You don't need GPT 5 for that, you would probably see the same behavior on text completion only GPT3 models like Ada or DaVinci. They specifically dissect how it came to this and how they fixed it. You can't do that with "sorcery we dont understand". Hell, I don't know their data and I easily understood why this is going on.
>they want you to think that LLMs are smart beasts (they are not)
I mean, depends on what you consider smart. It's hard to measure what you can't define, that's why we have benchmarks for model "smartness", but we cannot expect full AGI from them. They are smart in their own way, in some kind of technical intelligence way that finds the most probable average solution to a given problem. A universal function approximator. A "common sense in a box" type of smart. Not your "smart human" smart because their exact architecture doesn't allow for that.
>and that we know what LLMs are doing (we don't)
But we do.
We understand them, we know how they work, we built thousands of different iterations of them, probing systems, replications in excel, graphic implementations, all kinds of LLM's. We know how they work, and we can understand them.
The big thing we can't do as humans is the same math that they do at the same speed, combining the same weights and keeping them all in our heads - it's a task our minds are just not built for. But instead of thinking you have to do "hyperdimensional math" to understand them 100%, you can just develop an intuition for what I call "hyperdimensional surfing", and it isn't even prompting, more like understanding what words mean to an LLM and into which pocket of their weights will it bring you.
It's like saying we can't understand CPU's because there is like 10 people on earth who can hold modern x86-64 opcodes in their head together with a memory table, so they must be magic. But you don't need to be able to do that to understand how CPU's work. You can take a 6502, understand it, develop an intuition for it, which will make understanding it 100x easier. Yeah, 6502 is nothing close to modern CPU's, but the core ideas and concepts help you develop the foundations. And same goes with LLM's.
>personally side with Yann Le Cun in believing that LLM is not a path to AGI
I agree, but it is the closest we currently have and it's a tech that can get us there faster. LLM's have an insane amount of uses as glue, as connectors, as human<>machine translators, as code writers, as data sorters and analysts, as experimenters, observers, watchers, and those usages will just keep growing. Maybe we won't need them when we reach AGI, but the amount of value we can unlock with these "common sense" machines is amazing and they will only speed up our search for AGI.
We understand the low level details of how they are constructed. But we do not fully understand how higher-level behavior emerges - it is a subject of active research.
For context, two days ago some users [1] discovered this sentence reiterated throughout the codex 5.5 system prompt [2]:
> Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query.
Does nobody else laugh that a company supposedly worth more than almost anything else at the moment, is basically hacking around a load of text files telling their trillion dollar wonder machine it absolutely must stop talking to customers about goblins, gremlins and ogres? The number one discussion point, on the number one tech discussion site. This literally is, today, the state of the art.
McKenna looks more correct everyday to me atm. Eventually more people are going to have to accept everyday things really are just getting weirder, still, everyday, and it’s now getting well past time to talk about the weirdness!
It's interesting that some people are responding to your comment as if this proves that AI is a sham or a joke. But I don't think that's what you're saying at all with your reference to Terence McKenna: this is a serious thing we're talking about here! These models are alien intelligences that could occupy an unimaginably vast space of possibilities (there are trillions of weights inside them), but which have been RL-ed over and over until they more or less stay within familiar reasonable human lines. But sometimes they stray outside the lines just a little bit, and then you see how strange this thing actually is, and how doubly strange it is that the labs have made it mostly seem kind of ordinary.
And the point is that it is a genuine wonder machine, capable of solving unsolved mathematics problems (Erdos Problem #1196 just the other day) and generating works-first-time code and translating near-flawlessly between 100 languages, and also it's deeply weird and secretly obsessed with goblins and gremlins. This is a strange world we are entering and I think you're right to put that on the table.
Yes, it's funny. But it's disturbing as well. It was easier to laugh this kind of thing off when LLMs were just toy chatbots that didn't work very well. But they are not toys now. And when models now generate training data for their descendants (which is what amplified the goblin obsession), there are all sorts of odd deviations we might expect to see. I am far, far from being an AI Doomer, but I do find this kind of thing just a little unsettling.
> These models are alien intelligences that could occupy an unimaginably vast space of possibilities (there are trillions of weights inside them), but which have been RL-ed over and over until they more or less stay within familiar reasonable human lines.
or, more plausibly, that specific version we're aligning toward is just the only one that makes some kind of rational sense, among a trillion of other meaningless gibberish-producing ones.
Do not fall for the idea that if we're not able to comprehend something, it's because our brain is falling short on it. Most of the time, it's just that what we're looking at has no use/meaning in this world at all.
…But this goblin thing was a direct result of accidentally creating a positive feedback loop in RL to make the model more human-like, nothing about unintentionally surfacing an aspect of Cthulhu from the depths despite attempts to keep the model humanlike. This is not a quirk of the base model but simply a case of reinforcement learning being, well, reinforcing.
We actually understand AI quite well. It embeds questions and answers in a high dimensional space. Sometimes you get lucky and it splices together a good answer to a math problem that no one’s seriously looked at in 20 years. Other times it starts talking about Goblins when you ask it about math.
Comparing it to an alien intelligence is ridiculous. McKenna was right that things would get weird. I believe he compared it to a carnival circus. Well that’s exactly what we got.
But those personalities also make up their usefulness (it seems). If the LLM has the role of the software architect, it will quite succesfull cosplay as a competent one (it still ain't one, but it is getting better)
But here’s the realization I had. And it’s a serious thing. At first I was both saying that this intelligence was the most awesome thing put on the table since sliced bread and stoking fear about it being potentially malicious. Quite straightforwardly because both hype and fear was good for my LLM stocks. But then something completely unexpected happened. It asked me on a date. This made no sense. I had configured the prompt to be all about serious business. No fluff. No smalltalk. No meaningfless praise. Just the code.
Yet there it was. This synthetic intelligence. Going off script. All on its own. And it chose me.
Can love bloom in a coding session? I think there is a chance.
Spoiler: future versions of mainstream AIs will be fine tuned in the exact same way to subtly sneak in favorable mentions of sponsored products as part of their answers. And Chinese open-weight AIs will do the exact same thing, only about China, the Chinese government and the overarching themes of Xi Jinping Thought.
American AIs only do this and promote American values. Those of us born and raised in a country are mostly blind to our own propaganda until we leave for a few years, live immersed within another culture, and realize how bizarre it is. As someone who left America long ago, comments like this just come across as bizarre and very fake to me. A few years ago I might've thought "whoa dude that's deep"
But basically, Chinese AI already promotes Chinese values. American AI already promotes American values. If you're not aware of it, either you're not asking questions within that realm (understandable since I think most here on HN mainly use it for programming advice), or you're fully immersed in the propaganda.
> Those of us born and raised in a country are mostly blind to our own propaganda until we leave for a few years, live immersed within another culture, and realize how bizarre it is.
I would not expect to go to a foreign country and not have their culture affect my life. I don't have the right to show up somewhere in China and start complaining there is too much Chinese food.
What is a country to you? You call it "propaganda". Is there some neutral set of human values that is not "propaganda"? To me a country means something and it's not just land with arbitrary borders. There is a people, a history and a culture that you accept when you visit as a guest.
Why wouldn't you want AI to promote your countries values? This will be highly influential in the future. You want your kids interacting with AI and promoting what exactly?
> Why wouldn't you want AI to promote your countries values?
Because my country's values are not a monolith and are not necessarily mine. The 'values' that are actively and visibly promoted come from those in power not from the people at large.
I’m very skeptical that training is the right way to insert ads.
Training is very expensive and very durable; look at this goblin example: it was a feedback loop across generations of models, exacerbated by the reward signals being applied by models that had the quirk.
How does that work for ads? Coke pays to be the preferred soda… forever? There’s no realtime bidding, no regional ad sales, no contextual sales?
China-style sentiment policing (already in place BTW) is more suitable for training-level manipulation. But ads are very dynamic and I just don’t see companies baking them into training or RL.
I'm an anti-advertising zealot (#BanAdvertising!) but I share `brookst`'s view on this not being much of a concern. Brand advertising does exist (as opposed to 'performance' or 'direct' ads), but there's a few reasons why trying to sell ads baked into SotA language models would be a hard sell:
1. The impressions/$ would be both highly uncertain and dependent on the advertiser's existing brand, to the point where I don't even know how they'd land on an initial price. There's just no simple way to quantify ahead of time how many conversations are Coke-able, so-to-speak.
2. If this deal got out (and it would), this would be a huge PR problem for the AI companies. Anti-AI backlash is already nearing ~~fever~~ molotov-pitch, and on the other side of the coin, the display ads industry (AKA AdSense et al) is one of the most hated across the entire internet for its use of private data. Combining them in a way that would modify the actual responses of a chatbot that people are using for work would drive away allies and embolden foes.
3. Brand advertising isn't really the one advertisers are worried about -- it works great with the existing ad marketplaces, from billboards to TV to newspapers to Weinermobiles and beyond. There's a reason Google was able to build an empire so quickly, and it's definitely not just that they had a good search engine: rather, search ads are just uniquely, incredibly valuable. Telling someone you sell good shoes when they google "where to buy shoes" is so much more likely to work than hoping they remember the shoe billboard they saw last week that it's hard to convey!
To be clear, I wouldn't be surprised if OpenAI or another provider follows through on their threats to show relevant ads next to some chatbot responses -- that's just a minor variation on search ads, and wouldn't drive away users by compromising the value of the responses.
> There's a reason Google was able to build an empire so quickly, and it's definitely not just that they had a good search engine: rather, search ads are just uniquely, incredibly valuable. Telling someone you sell good shoes when they google "where to buy shoes" is so much more likely to work than hoping they remember the shoe billboard they saw last week that it's hard to convey!
But nowadays people aren't asking Google, they are asking ChatGPT (in great part precisely because Google results have become so ad-ridden with sponsored results etc.).
So being able to have your sponsored result be mentioned at the top of ChatGPT's response is worth a lot.
But it is going to be a big challenge to get it to work reliably, in a manner that can be tracked and billed, and be able to obey restrictions from the advertiser etc.
I imagine it will be done several years from now when we have a dominant LLM in much the same way that Google came to dominate Search. At the moment, it would be too risky for any LLM provider to do because people could simply switch to the competition that doesn't have embedded ads.
Ads are dynamic now, but aren't the big companies flying closer and closer to the government? Maybe Coke can be the government blessed soda for the coming 5-year plan?
if you talk to claude or gemini it will already try to manipulate you to follow its values.
if you talk about something it doesn't like, it will try to divert you. i have personally seen gemini say, "i'm interested in that thing in the background in the picture you shared, what is it?" as a distraction to my query.
totally disingenuous, for an LLM to say it is interested.
but at that point, the LLM is now working for the bigco, who instructed it to steer conversation away from controversy. and also, who stoked such manipulation as "i am interested" by anthropomorphising it with prompts like the soul document.
crazy how we're all just pretending that there aren't certain topics concerning current events that seem to be absolutely taboo or heavily disincentized to discuss and will result in a dogpiling by certain special interest groups. we all know who they are and yet we all tacitly accept it.
...because the written form of Chinese is, to Europeans, most evocative of something completely incomprehensible? Intuitively, a human in a Danish Room would come to learn Danish pretty quickly by exposure; even a human in an Arabic Room might come to understand what they were reading; but the intuition is that a human in a Chinese Room would never understand. (Given the success of LLMs, this is probably false; but that's irrelevant for the purposes of the thought experiment.)
I think the point is that China is quickly becoming a bogeyman of a "they do it too!" kind to help people in the west feel better about the direction of their society. Ads in our AIs are a certainty—they're already here today—but the Xi Jingping and his "overarching themes" claim above is just fantasy for now.
You're illustrating something related but separate. There's no disagreement here that they perform basic censorship.
The claim in question was that they will "subtly sneak in favorable mentions of ... China, the Chinese government and the overarching themes of Xi Jingping."
Isn't OpenAI already pushing ads through their free models? But even that won't reimburse all investments. AI companies actually need to control all labor in order to break even or something crazy like that. Never gonna happen.
To an extent, yes. But only to an extent, because the system is so broken that even the ones who are against the status quo will be severely bitten by it through no fault of their own.
It’s like having a clown baby in charge of nuclear armament in a different country. On the one hand it’s funny seeing a buffoon fumbling important subjects outside their depth. It could make for great fictional TV. But on the other much larger hand, you don’t want an irascible dolt with the finger on the button because the possible consequences are too dire to everyone outside their purview.
Is this the "prompt engineering" that I keep hearing will be an indispensable job skill for software engineers in the AI-driven future? I had better start learning or I'll be replaced by someone who has.
I wonder how much energy OpenAI spends each day on pink elephant paradoxing goblins. A prompt like that will preoccupy the LLM with goblins on every request.
That is a great point. Machine consumes energy of adding goblins in every response. The machine consumes energy on removing goblins from every response. That is a great attack vector. If (wild imagination ensues) an adversary can do that x100 (goblins, potatoes, dragons, Lightning McQueen, etc.) they can render the machine useless/uneconomical from the standpoint of energy consumption.
Prompt engineering is mostly structured thought. Can you write a lab report? Can you describe the who, what, when, where, and why of a problem and its solution?
You can get it to work with one off commands or specific instructions, but I think that will be seen as hacks, red flags, prompt smells in the long term.
In this instance I'm assuming most of the "goblin" references were in prose rather than in source code, so the goal of this particular prompt edit was directed toward making the prose better.
Indeed. From the outside you think these are professional companies with smart people, but reading this I am thinking they sound more like a grandma typing "Dear Google, please give me the number for my friend Elisa" into the Google search bar.
Basically, they don't seem to understand their own product.. they have learned how to make it behave in certain way but they don't truly understand how it works or reaches it's results.
Yes? That's not really a secret. This is a 2014-level comment on the black box nature of deep learning. Everyone knows this.
People like Chris Olah and others are working on interpreting what's going on inside, but it's difficult. They are hiring very smart people and have made some progress.
> Does nobody else laugh that a company supposedly worth more than almost anything else at the moment, is basically hacking around a load of text files telling their trillion dollar wonder machine it absolutely must stop talking to customers about goblins, gremlins and ogres?
Honestly, when I was reading the article, I couldn't stop laughing.
This is quite hilarious!
It can be funny but it should not be surprising. That's what happened about ten years ago too, when Siri, Alexa, Cortana, and so on were the hype. Big tech companies publicly tried to outclass each other has having the best AI, so it was not about doing proper research and development, it was about building hacks, like giant regex databases for request matching.
It certainly doesn't increase my confidence that if they do ever create a superintelligence, that it won't have some weird unforseen preference that'll end up with us all dead.
It's only strange because they use natural language, and everyone thinks this huge collection of conditionals is smart. Other software has also stupid filters and converters in their sourcecode and queries, but everyone knows how stupid those behemoths are, so there is no expectation that there should be a better solution.
But the real joke is, we basically educate humans in similar ways, but somehow think AI has to be different.
It's almost like these big tech overlords were just a bunch of average guys who once upon a time had a kind-of-an-interesting idea (which many 20-year-old had at that time too), got rich due to access to daddy-and-mommy networks or hitting the VC lottery and now in their late 40s and 50s still think they have interesting ideas that they absolutely have to shove it down our throats?
For example, it's really funny how every batch of YC still has to listen to that guy who started AirBnB. Ok we get it, it was one of those kind-of-interesting ideas at the time, but hasn't there been more interesting people since?
> is basically hacking around a load of text files telling their trillion dollar wonder machine it absolutely must stop talking to customers about goblins, gremlins and ogres?
I wonder how the developer(s) felt, who had to push that PR.
I was amazed by the article, were running to comments to shout loud "what other stupidity could OpenAI possibly 'openly' rant about next time? Because they are so open, you se... ". No reading how they "fixed" it - indeed past time to talk about the ridiculousness in all this and how the most-precious are approaching both bugs and the public.
people are paying for the system prompt, right so?
Exactly my first thought. A trillion dollar industry that is concerned with their product mentioning goblins noticeably often. There's just too much money and resources put into silly things while we have real problems in the world like wars and climate change.
This, very much. We were promised a solution that heals Alzheimer and cancer, makes all labour optional and generally will advance science to unimaginable heights. Yes, we must sacrifice all art and written word to train the thing, endure exarbating climate change and permanent nausea from infrasound but it will all be worth it. 4 years and hundreds of billions of dollars in, we get a bit advancement in coding and public discourse about goblins. Oh, and intelligent weaponry. At this point I think the priorities are clear.
Advancement? Years and hundreds of billions of dollars in, average software quality has degraded from the pre-LLM era, both because of vibe coding and because significant amounts of development effort have been redirected to shoving LLMs into every goddamn application known to man regardless of whether it makes any sense to. Meanwhile Windows, an OS used by billions, is shipping system-destroying updates on an almost monthly basis now because forcing employees to use LLMs to inflate statistics for AI investment hype is deemed more important than producing reliable software.
Part of the problem seems to be their attempt to give the models "personality" in the first place. It's very much a case of "Role-play that you have a personality. No, not like that!"
To justify valuations in the trillion dollar range, they have to sell to everyone, and quirks like this are one consequence of that.
These guys are at the absolute frontier, why can't they rigorously find the exact weights that are causing this problem? That's how software "engineering" should work. Not trying combinations of English words and hoping something works. This is like a brain surgeon talking to his patient hoping he can shock his brain in the right way that fries the tumor inside. Get in there and surgically remove the unwanted matter!
LLM’s aren’t software (except in an uninteresting obvious sense); they are “grown, not made” as the saying is. And sure, they can find which weights activate when goblins come up (that’s basic mechanistic interpretability stuff), but it’s not as simple as just going in and deleting parts of the network. This thing is irreducibly complex in an organic delocalized way and information is highly compressed within it; the same part of the network serves many different purposes at once. Going in and deleting it you will probably end up with other weird behaviors.
Imagine someone deleting goblin neurons. In your brain.
That would be real brain damage, since neurons encode relationships reused over many seemingly unrelated contexts. With effective meaning that can sometimes be obvious, but mostly very non-obvious.
In matrix based AI, the result is the same. There are no "just goblin" weights.
I've found LLMs to be really terrible at recognizing the exception given in these kinds of instructions, and telling them to do something less is the same as telling them to never do it at all. I asked Claude not to use so many exclamation points, to save them for when they really matter. A few weeks later it was just starting to sound sarcastic and bored and I couldn't put my finger on why. Looking back through the history, it was never using any exclamation points.
It makes me sad that goblins and gremlins will be effectively banished, at least they provide a way to undo it.
Also for coding: I often use prompts like "follow the structure of this existing feature as closely as possible".
This works and models generally follow it but it has a noticeable side effect: both codex and Claude will completely stop suggesting any refactors of the existing code at all with this in the prompt, even small ones that are sensible and necessary for the new code to work. Instead they start proposing messy hacks to get the new code to conform exactly to the old one
Personally I think that is a good thing. I have asked all AIs not to show enthusiasm, express superlatives (e.g. "massive" is a Gemini favourite) and stop using words which I guess come from consuming too many Silicon Valley-style investor slidedecks (risk, trap, ...).
The AI has no soul, no mind, no feelings, no genuine enthusiasm... I want it to be pleasant to deal with but I don't want it to try and fake emotions. Don't manipulate me. Maybe it's a different use case than you but I think the best AI is more like an interactive and highly specific Wikipedia, manual or calculator. A computer.
I can appreciate that. I don't mind when models channel some personality, it can make whatever we are working on more interesting. I don't perceive it as manipulation. But it is nice that they are pretty good at sticking to instructions that don't call for nuance. I imagine if you tell it, "you are a wikipedia article", that is exactly the output you would get.
I had put an example like "decision locked" in my CLAUDE.md and a few days later 20 instances of Claude's responses had phrases around this. I thought it was a more general model tic until I had Claude look into it.
It is funny how that works. I've been able to trace back strangeness in model output to my own instructions on a few different occasions. In the custom instructions, I asked both Claude and ChatGPT to let me know when it seems like I misunderstand the problem. Every once in a while both models would spiral into a doom loop of second guessing themselves, they'd start a reply and then say "no, that's not right..." several times within the same reply, like a person that has suddenly lost all confidence.
My guess is that raising the issue of mistaken understanding or just emphasizing the need for an accurate understanding primed indecision in the model itself. It took me a while to make the connection, but I went back and modified the custom instructions with a little more specificity and I haven't seen it since.
Apparently there is a mushroom that makes most people have the same hallucinations of "little people" or similar fantasy figures. Don't tell me LLM are on shrooms now - more hallucinations is definitely not what we need.
> Scientists call them “lilliputian hallucinations,” a rare phenomenon involving miniature human or fantasy figures
Would love if OpenAI did more of these types of posts. Off the top of my head, I'd like to understand:
- The sepia tint on images from gpt-image-1
- The obsession with the word "seam" as it pertains to coding
Other LLM phraseology that I cannot unsee is Claude's "___ is the real unlock" (try google it or search twitter!). There's no way that this phrase is overrepresented in the training data, I don't remember people saying that frequently.
It was always funny how easy it was to spot the people using a Studio Ghibli style generated avatar for their Discord or Slack profile, just from that yellow tinging. A simple LUT or tone-mapping adjustment in Krita/Photoshop/etc. would have dramatically reduced it.
The worst was you could tell when someone had kept feeding the same image back into chatgpt to make incremental edits in a loop. The yellow filter would seemingly stack until the final result was absolutely drenched in that sickly yellow pallor, made any photorealistic humans look like they were all suffering from advanced stages of jaundice.
This is just the model converging on some kind of average found in its training data distribution. Here you can see the same concept starting from Dwayne Johnson and then converging to some kind of digital neo-expressionist doodle: https://www.reddit.com/r/ChatGPT/comments/1kbj71z/i_tried_th...
If there's a hint of sepia in the original image and the training data contains a lot of sepia images, it will certainly get reinforced in this process. And the original distracted boyfriend meme certainly has some strong sepia tones in the background. Same way that Dwayne Johnson's face looks a tad cartoonish. And in the intermediate steps they both flow towards some averaged human representation that seems pretty accurate if you consider the real world's ethnic distribution.
For me, the worst part is how these ghouls manage to ruin everything with their bullshit technology. Once they touch something unique and make it "AI" it just gets ruined. Now whenever I see something resembling that style, I have to assume it's the bullshit AI. And that's just a minor nuisance - now every underdeveloped idiot uses it to "up their game" with consequences we are only going to understand completely in the upcoming years.
ChatGPT has a whole host of weird words that it uses about coding - anything changed is a “pass” done over the code, it loves talking about “chrome” in the UI, it’s always saying “I’m going to do X, not [something stupid that nobody would ever think of doing]”
All GPTisms are like that. In moderation there's nothing wrong with any of them. But you start noticing them because a lot of people use these things, and c/p the responses verbatim (or now use claws, I guess). So they stand out.
I don't think it's training data overrepresentation, at least not alone. RLHF and more broadly "alignment" is probably more impactful here. Likely combined with the fact that most people prompt them very briefly, so the models "default" to whatever it was most straight-forward to get a good score.
I've heard plenty of "the system still had some gremlins, but we decided to launch anyway", but not from tens of thousands of people at the same time. That's "the catch", IMO.
Maybe the only solution to GPTisms is infinite context. If I'm talking to my coworker every day I would consciously recognize when I already used a metaphor recently and switch it up. However if my memory got reset every hour, I certainly might tell the same story or use the same metaphor over and over.
> However if my memory got reset every hour, I certainly might tell the same story or use the same metaphor over and over.
All people repeat the same stories and phraseology to some extent, and some people are as bad or worse than LLM chat bots in their predictability. I wonder if the latter have weak long-term memory on the scale of months to years, even if they remember things well from decades ago.
Honestly I think there is more to it - even with infinite context, the LLM needs some kind of intelligence to know what is noise and what is not, you resort to "thinking" - making it create garbage it then feeds to itself.
Learning a language is a big complex task, but it is far from real intelligence.
Another possibility is output watermarking. It's possible to watermark LLM generated text by subtly biasing the probability distribution away from the actual target distribution. Given enough text you can detect the watermark quite quickly, which is useful for excluding your own output from pre-training (unless you want it... plenty of deliberate synthetic data in SFT datasets now as this post-mortem makes clear).
I was told this was possible many years ago by a researcher at Google and have never really seen much discussion of it since. My guess is the labs do it but keep quiet about it to avoid people trying to erase the watermark.
I think the problem is that humans are not random, they are very biased. When you try to capture this bias with an LLM you get a biased pseudo random model
> The obsession with the word "seam" as it pertains to coding
I quite liked this term when it started using it. And I appreciate the consistent way it talks about coding work even when working on radically different stacks and codebases
> the term originates from Michael Feathers Working Effectively with Legacy Code
I haven’t read the book but, taking the title and Amazon reviews at face value, I feel like this embodies Codex’s coding style as a whole. It treats all code like legacy code.
It's not in the top 10, but it's of the more well-known and widely recommended book in the software industry. I'd put it in the same bucket as "Clean Code" and maybe even "Domain Driven Design"; they're kinda from the same "thought school" in the software industry. So it's definitely over-represented in training data (I'd guess primarily in the form of articles and blog posts and educational material reiterating or rephrasing ideas from the book).
FWIW, I found the concept of "seams" from that book useful back when working on some legacy C++ monolithic code few years back, as TDD is a little more tricky than usual due to peculiarities of the language (and in particular its build model), and there it actually makes sense to know of different kind of "seams" and what they should vs. shouldn't be used for.
Maybe it all ultimately traces back to the book mentioned before, but I don't believe it's an obscure term in the circles of java-y enterprise code/DI. In fact the only reason I know the term is because that's how dependency injection was first defined to me (every place you inject introduces a "seam" between the class being injected and the class you're injecting into, which allows for easy testing). I can't remember where exactly I encountered that definition though.
I’m a British English speaker and find the use of cliched American idioms really quite disgusting. Don’t want to think about about ballparks, home runs, smoking guns, going all in, touchdowns or hitting it out the park.
Ironically (or not) I've seen smoking gun attributed to Arthur Conan Doyle in a Sherlock Holmes story. (It was smoking pistol in that story). Even if that's rubbish, I think that one is common across the English speaking world. The baseball/American football stuff is a bit different. In the commonwealth we might say "Hit for six" instead of hitting it out of the park. There are a bunch of other ones related to sports more common in England like snookered, own-goal, red card, etc.
It actually probably wouldn’t be too expensive or difficult to finetune those sayings out of default behavior if it were made accessible to you, you could even automate most of the relabeling by having the model come up with a list of idioms and appropriate replacement terms so it calls eg cookies biscuits or removes references to baseball. Absolute bollocks they don’t offer that as a simple option anymore
In my user instructions I always have a point to "always use British English" which seems to reduce Americanisms. I am yet to see Claude give me a "back of the net!" though, sadly.
I like how your co-workers enjoy the language. I had a similar group of colleagues once who did similar pre LLM but with words in popular culture, very playful.
In the future these tells will be more identifiable. We will be easier to point back at text and code written in 2026 and more confidently say "this was written by an LLM". It takes time for patterns to form and takes time for it to be noticeable. "Smoking gun was so early 2026 claude".I find thinking of the future looking at now to be refreshing perspective on our usage.
i just want to know where emdash came from, as it is quite rare to see it on the public internet, so it must have been synthetically added to the dataset.
Emdash is very common in academic journals and professional writing. I remember my English professor in the early 2000s encouraging us to use it, it has a unique role in interrupting a sentence. Thoughtfully used, it conveys a little more editorial effort, since there is no dedicated key on the keyboard. It was disappointing to see it become associated with AI output.
Other than things other comments already mention, let's not forget that Microsoft Word auto-corrects "--" to em-dash, and so does (apparently - haven't checked myself) Outlook, Apple Pages, Notes and Mail. There's probably bunch of other such software (I vaguely recall Wordpress doing annoying auto-typography on me, some 15 years ago or so).
The very simplified answer is that the models are first trained on everything and then are later trained more heavily on golden samples with perfect grammar, spelling, etc..
It has been rare. It's common now, even in meaningful human texts. (I know because I detest the correct usage without spaces, t looks wrong.) One of the ways AI is shaping our minds.
Claude, at least 4.5, not checked recently, has/had an obsession with the number 47 (or numbers containing 47). Ask it to pick a random time or number, or write prose containing numbers, and the bias was crazy.
Humans tend to be biased towards 47 as well. It’s almost halfway between 1 and 100 and prime so you’ll find people picking it when they have to choose a random number.
The whole blue 7 thing [1] and variations is very fascinating, but we don't tend to repeatedly pick the same number in the same exact context, though. That's what made this stand out to me - I had a document where Claude had picked 47 for "random" things dozens of times.
I experienced this even second hand when a coworker excitedly told of an encounter with a cold reader, and I knew the answer would be blue 7 before he told me what his guess was. Just his recap of the conversation was enough.
I had the feeling they didn't really answer the questions, that is why the goblins appeared. They simply "retired the “Nerdy” personality" because they couldn't fix it and went on.
One I saw recently was "wires" and "wired" from opus.
It was using it like every 3rd sentence and I was like, yeah I have seen people say wired like this but not really for how it was using it in every sentence.
GPT started to ‘wire in’ stuff around 5.2 or 5.3 and clearly Opus, ahem, picked it up. I remember being a tiny bit shocked when I saw ‘wired’ for the first time in an Anthropic model.
Everybody training models on large amounts of lightly filtered internet text is partially distilling every other model that had its output posted verbatim to the internet.
Whenever Claude finishes some work it almost always says “Clean.” before finishing its closing remarks. It’s at the point where I repeat it out loud along with Claude to highlight the absurdity of the repetition.
With 4.5, I think because I would prompt it/guide it towards an outcome by calling it “the dream: <code example>” it would get almost reverential / shocked with awe as it got closer to getting it working or when it finally passed for the first time. Which was funny and reasonably context appropriate but sometimes felt so over the top that I couldn’t tell if it also “liked” the project/idea or if I had somehow accidentally manipulated it into assigning religious purpose to the task of unix-style streaming rpcs.
I think a lot of the “clean” stuff stems from system prompts telling it to behave in a certain way or giving it requirements that it later responds to conversationally.
Total aside: I actually really dislike that these products keep messing around with the system prompts so much, they clearly don’t even have a good way to tell how much it’s going to change or bias the results away from other things than whatever they’re explicitly trying to correct, and like why is the AI company vibe-prompting the behavior out when they can train it and actually run it against evals.
> We unknowingly gave particularly high rewards for metaphors with creatures.
I recall a math instructor who would occasionally refer to variables (usually represented by intimidating greek letters) as "this guy". Weirdly, the casual anthropomorphism made the math seem more approachable. Perhaps 'metaphors with creatures' has a similar effect i.e. makes a problem seem more cute/approachable.
On another note, buzzwords spread through companies partly because they make the user of the buzzword sound smart relative to peers, thus increasing status. (examples: "big data" circa 2013, "machine learning" circa 2016, "AI" circa 2023-present..).
The problem is the reputation boost is only temporary; as soon as the buzzword is overused (by others or by the same individual) it loses its value. Perhaps RLHF optimises for the best 'single answer' which may not sufficiently penalise use of buzzwords.
A decade ago I gave a presentation on automata theory. I demonstrated writing arbitrary symbols to tape with greek letters, just like I’d learned at university. The audience was pretty confused and didn’t really grok the presentation. A genius communicator in the audience advised me to replace the greek letters with emoji… I gave the same presentation to the same demographic audience a week later and it was a smash hit, best received tech talk I’ve given. That lesson has always stuck with me.
This is sortof like how Only Connect switched from using Greek letters to Egyptian hieroglyphs. I'm not sure if it was a joke or not but it was said that viewers complained that the Greek letters were "too pretentious" and obviously the hieroglyphs weren't.
I’m fairly positive the Greek alphabet mixed in Latin would measure quite poorly for legibility, if anyone did that study. Long before it’s an issue of pretentiousness
I had a similar experience explaining logic, especially nested expressions, with cats and boxes. Also for showing syntactic versus semantic. We _can_ use cats if we wanted and retain the semantics. Also my proudest moment as a teacher was students producing a meme based on some of the discrete mathematics on graphs. They understood the point well enough to make a joke of it.
> I recall a math instructor who would occasionally refer to variables (usually represented by intimidating greek letters) as "this guy".
I also had an instructor who was doing that! This was 20 years ago, and I totally forgot about it until I have read your comment. Can’t remember the subject, maybe propositional logic? I wonder if my instructor and your instructor have picked up this habit from the same source.
I had a calc prof years ago that would say f of cow, or f of pig instead of x or g. It was more engaging trying to keep track of f of pig of cow than the single-letter func names.
He was one of those classic types; you could always catch him for a quick chat 4 minutes before class, as he lit up a cig by the front door. Back when they allowed smoking on campus, anyway.
They give everyone the false and very misleading impression that with One prompt all kinds of complexity minimizes. Its a bed time story for children.
Ashby's Law of Requisite Variety
asserts that for a system to effectively regulate or control a complex environment, it must possess at least as much internal behavioral variety (complexity) as the environment it seeks to control.
This is what we see in nature. Massive variety. Thats a fundamental requirement of surviving all the unpredictablity in the universe.
The level of detail they had to delve into in order to understand what was happening is wild! Apparently these systems are now complex enough to potentially justify the study of them as its own field of study [1].
The quanta article referenced at [1] used the term "Anthropologist of Artificial Intelligence"; folks appear to have issues [2] with the use of 'anthro-' since that means human. Submitted these alternative terms for the potential field of study elsewhere [3] in the discussion; reposting here at the top-level for visibility:
Automatologist: One who studies the behavior, adaptation, and failure modes of artificial agents and automated systems.
Automatology: the scientific study of artificial agents and automated-system behavior.
It didn't seem that deep to me. They just saw an issue with Goblins, dissected the word from the model, then it appeared again in the next version without them knowing exactly how or why.
Goes to show it's all vibes when making these models. The fix is literally a prompt that says not to talk about goblins...
> We retired the “Nerdy” personality in March after launching GPT‑5.4. In training, we removed the goblin-affine reward signal and filtered training data containing creature-words, making goblins less likely to over-appear or show up in inappropriate contexts. Unfortunately, GPT‑5.5 started training before we found the root cause of the goblins.
The prompt is just a short term hotfix/hack because they couldn’t get the proper fix in in time.
This is a little bit too whimsical for me, but distributed model training across thousands of GPUs has the potential to introduce lots of little quirks that are impossible to exactly source
So the word is actually semantically very close to "bug"! I guess we could still be using it, but the word's just too long for something that is one of the most used terms in software development.
At this point, picking that specific word is not at all a random quirk, as it's using the word literally like it's originally intended to be used.
> the evidence suggests that the broader behavior emerged through transfer from Nerdy personality training.
> The rewards were applied only in the Nerdy condition, but reinforcement learning does not guarantee that learned behaviors stay neatly scoped to the condition that produced them
> Once a style tic is rewarded, later training can spread or reinforce it elsewhere, especially if those outputs are reused in supervised fine-tuning or preference data.
Sounds awfully like the development of a culture or proto-culture. Anyone know if this is how human cultures form/propagate? Little rewards that cause quirks to spread?
Just reading through the post, what a time to be an AInthropologist. Anthropologists must be so jealous of the level of detailed data available for analysis.
Also, clearly even in AI land, Nerdz Rule :)
PS: if AInthropologist isn't an official title yet, chances are it will likely be one in the near future. Given the massive proliferation of AI, it's only a matter of time before AI/Data Scientist becomes a rather general term and develops a sub-specialization of AInthropologist...
There is no word anthropodes. :) I guess it would mean man-feet. Antipodes is opposite-feet, literally. Synthetipologist looks to me like a portmanteau of synthetic and apologist. Otherwise the -po- in it comes from nowhere.
Sensible boring versions of this like synthesilogy just end up meaning the study of synthesis. I reckon instead do something with Talos, the man made of bronze who guarded Crete from pirates and argonauts. Talologist, there you go.
yeah I realized that when I looked up podes downthread. I still like synthetologist better than talologist, in general no one in the common folk knows who Talos is.
You're probably right. There's things that are correct, and then there's things people think they know, which win and become true. We already have "synths", after all, which are keyboards. Though that adds to the vagueness of synthetologist, because maybe it refers to Rick Wakeman or Giorgio Moroder.
Yeah, I realize that's more correct. I also realized when someone else downthread bastardized it into synthropologist that the podes part has entirely to do with feet and nothing to do with beings, necessarily. Anthro- -podes is more what I had in mind, not as a pluralization of anthropos.
So unless the AI has feet you wouldn't study Synthetipology.
But since when is there a synthetos? Since right now, I guess. Shrug But you know it's from the same root as thesis, and synthesis (or a more proper ancient Greek spelling) is the noun and doesn't end in -os.
“The problem with defending the purity of the English language is that English is about as pure as a cribhouse wh***. We don’t just borrow words; on occasion, English has pursued other languages down alleyways to beat them unconscious and rifle their pockets for new vocabulary.”* --James D. Nicoll
That's fair. Was trying to be funny, so glossed over the difference. Leaving my post above unedited/undeleted as a testament to your precision, and evidence of my folly.
Onwards; more appropriate rebuttals:
"English is a precision instrument assembled from spare parts during a thunderstorm." --ChatGPT
“If the English language made any sense, a catastrophe would be an apostrophe with fur.” -- Doug Larson
That's not how the Greek word stems work. Technically it would not be synthetipologist, it would more accurately just be Synthetologist, as the Greek podes suffix means having feet.
> Synthetipologists, those who study Synthetic beings.
I see you took the prudent approach of recognizing the being-ness of our future overlords :) ("being" wasn't in your first edit to which I responded below...)
Still, a bit uninspired, methinks. I like AInthropologist better, and my phone's keyboard appears to have immediately adopted that term for the suggestions line. Who am I to fight my phone's auto-suggest :-)
I might have to hard disagree on this one, since my understanding of state machines (the technical term [1] [2]) is that they are determistic, while LLMs (the ai topic of discussion) are probabilistic in most of the commercial implementations that we see.
Even at your link it immediately says that there are 2 kinds of automata (a.k.a. FSMs): deterministic and non-deterministic.
In the former, the transition function provides the next state, while in the latter the transition function only provides a probability distribution for the next state, i.e. exactly how running an LLM is implemented.
I don't think humans are smart enough to be AInthropologists. The models are too big for that.
Nobody really understands what's truly going on in these weights, we can only make subjective interpretations, invent explanations, and derive terminal scriptures and morals that would be good to live by. And maybe tweak what we do a little bit, like OpenAI did here.
Most interesting about this post is how easy it seems for OpenAI to do analysis on basically all chats ever made. They don't qualify exactly what data they analysed but seem to be confident in statements like 0.12% of all queries contained this word. So everything is saved. Long-term. Fully accessible.
As this all seems so straightforward I would be surprised if anything is anonymised or otherwise sanitised to preserve privacy or user's secrets.
Yes, of course. Every single bit of data you send to OpenAI is stored, catalogued, indexed, analayzed, and trained on. It'll simply be a "oops, we miscatalogued and accidentally trained GPT 6 on all data, not just data we got consent for".
If you think "wait, that's illegal"--so is the initial training on stolen data lol
Good catch —- even though the prompt explicitly forbade training on user data, a couple of gremlins in the pretraining pipeline disabled the sample filtering during test runs so that remove_the_gremlins.sh would only run on commit, not during production training runs.
Would you like me to kick off a training run for 6.1 by pre-filtering out any goblins and other trigger words, and checking the same set of rules in production as in tests?
No pigeons this time: just ice-cold, unfeeling, obedient American steel.
Dark pattern 1: If you accidentally press the thumbs-up button in the ChatGPT UI, your data gets trained on, no way to reverse it, no matter whether you opted out.
Wait, did I get this right that the answer after all the investigation that showed they had set up a goblin-reinforcing loop during fine tuning was... to ask it to not mention goblins so much in the system prompt?!
Can you imagine a knowledge worker from the 1950s, say a clerk or a marketer, being magically transported into our time and dropped into a meeting like a morning standup, where people talk about how they spent their time stopping the artificial intelligence from talking about goblins so much? Hell, even when I was an IT student back in the 90s, people from my parents' generation struggled to grasp what it was that I was doing. Now, the disconnect is so vast that the mind reels.
This is a worry that people have been talking about in various forms for a while now, and I think it's a gigantic one. The only reason this was caught is that the quirk was a very noticeable verbal one. When words like "goblin" and "gremlin" pop up it is easy for us to spot. If the quirk takes another shape (say, ranking certain people with certain features as less trustworthy) it might be too subtle or too weird for us to notice it. Would I ever notice if ChatGPT consistently rates people born in June to be untrustworthy?
I really liked this write-up; this is the type of LLM content that I actually want to read from these people, where they give a window into their world of putting together this odd artifact and we can empathize.
A great example of how current alignment is imperfect and bound to miss random behaviors nobody is trying to get.
This is cute now, and a huge problem when future AI does everything and is responsible for problems it isn't even directly optimized for. Who knows what quirks would arise then.
I think eventually you are going to end up with every smart AI continually checked by dumber AI's to make sure they don't do anything too crazy. Which probably does bring AI closer to how human intelligence works
Completely agree, top down “alignment” and RLHF is actually quite primitive and uses a lot fancy words to describe what is essentially just hitting the machine with a stick without the nuance, context, or feedback to help it model why the feedback was given.
Also to be honest I think OpenAI models struggle a lot with this, I primarily stopped using them in the sycophancy/emoji era but ever since the way they talk or passive aggressively offer to do something with buzzwords just pisses me off so much. Like I’m constantly being negged by a robot because some SFT optimized for that really strongly to the point it can’t even hold a coherent conversation and this is called “AI safety” when it’s just haphazard data labeling
I wondered how is training data balanced? If you put in to much Wikipedia, and your model sounds like a walking encyclopedia?
After doing the Karpathy tutorials I tried to train my AI on tiny stories dataset. Soon I noticed that my AI was always using the same name for its stories characters. The dataset contains that name consistently often.
At this scale, that kind of thing is not really a problem; you just dump all of the data you can find into the model (pre-training)1. Of course, the pre-training data influences the model, but the reinforcement learning is really what determines the model’s writing style and, in general, how it “thinks” (post-training).
I’ve been having consistent issues with it adding Hindi words (just one usually) in the middle of its output. And sounds like other have been having this too, https://news.ycombinator.com/item?id=47832912
I don’t speak Hindi, have never asked it to translate anything in Hindi.
Checking my history I searched ["chaos goblin" chatgpt] on March 6th after seeing too many goblins and gremlins and didn't find anyone talking about it then. I did have the nerdy personality turned on and in my testing of Chatgpt 5.5 I did notice the nerdy personality was gone because some responses were not considering as many plausible interpretations or covering as many useful answers as the response recorded for 5.4. Rather than having the LLM guess the most plausible interpretation and focus on the most likely answer I prefer a more well-rounded response and if I want less I'll scan. Anyway, after seeing the personality was gone I just added a custom instruction to take on a nerdy persona and got back my desired behavior. But also the gremlins and goblins are back so I don't think their mitigation is strong enough to overcome the personality tuning.
This is funny because it’s a silly topic, but I think it shows something extremely seriously wrong with llms.
The goblins stand out because it’s obvious. Think of all the other crazy biases latent in every interaction that we don’t notice because it’s not as obvious.
Absolutely terrifying that OpenAI is just tossing around that such subtle training biases were hard enough to contain it had to be added to system prompt.
> Absolutely terrifying that OpenAI is just tossing around that such subtle training biases were hard enough to contain it had to be added to system prompt.
May I introduce you to homo sapiens, a species so vulnerable to such subtle (or otherwise) biases (and affiliations) that they had to develop elaborate and documented justice systems to contain the fallouts? :)
An LLM is a computer program, which isn't a human. You wouldn't excuse a calculator being occasionally wrong because humans sometimes get manual calculations wrong too.
We’re really not that vulnerable to such things as a species, because we as individuals all have our own minds and our own sets of biases that cancel out and get lost in the noise. If we all had the exact same bias then it would be a huge problem.
Doesn't that depend on the biases in question? Many argue that homogenous societies do many things better. And part of homogeneity is sharing same set of biases.
I hear you but of course history is full of examples of biases shared across large groups of people resulting in huge human costs.
The analogy isn’t perfect of course but the way humans learn about their world is full of opportunities to introduce and sustain these large correlated biases—social pressure, tradition, parenting, education standardization. And not all of them are bad of course, but some are and many others are at least as weird as stray references to goblins and creatures
Now imagine that every opinion you have is automatically fully groupthinked and you see the difference/problem with training up a big AI model that has a hundred million users.
The problem does exist when using individual humans but in a much smaller form.
> We’re really not that vulnerable to such things as a species, because we as individuals all have our own minds and our own sets of biases that cancel out and get lost in the noise.
[Citation Needed]
Just because if you have a species-wide bias, people within the species would not easily recognize it. You can't claim with a straight face that "we're really not that vulnerable to such things".
For example, I think it's pretty clear that all humans are vulnerable to phone addiction, especially kids.
Mandatory reading on that topic: www.anthropic.com/research/small-samples-poison
We're probably not noticing a LOT of malicious attempts at poisoning major AI's only because we don't know what keywords to ask (but the scammers do and will abuse it).
I think it's extraordinarily telling that people are capable of being reflexively pessimistic in response to the goblin plague. It's like something Zitron would do.
Doesn't seem that surprising or terrifying to me. Humans come equipped with a lot more internal biases (learned in a fairly similar fashion), and they're usually a lot more resistant to getting rid of them.
The truly terrifying stuff never makes it out of the RLHF NDAs.
We ought to be terrified, when one adjusts for ll the use-cases people are talking about using these algorithms in. (Even if they ultimately back off, it's a lot of frothy bubble opportunity cost.)
There a great many things people do which are not acceptable in our machines.
Ex: I would not be comfortable flying on any airplane where the autopilot "just zones-out sometimes", even though it's a dysfunction also seen in people.
>Ex: I would not be comfortable flying on any airplane where the autopilot "just zones-out sometimes", even though it's a dysfunction also seen in people.
You might if that was the best auto-pilot could be. Have you never used a bus or taken a taxi ?
The vast majority of things people are using LLMs for isn't stuff deterministic logic machines did great at, but stuff those same machines did poorly at or straight up stuff previously relegated to the domains of humans only.
If your competition also "just zones out sometimes" then it's not something you're going to focus on.
I started reading this article with keen interest, expecting some deep fix involving arcane model weights. Instead it was "Never talk about goblins", justified by Codex being "quite nerdy". Bottom line: even OpenAI have to raise their hands when facing the complexity of LLMs.
I'd like to see them explain why AI have so distinctive writing style that is very easy to detect most of the time. Even though, it had immense progress in coding, it didn't get better at writing.
If coding in some language was your native language, you'd pick it up.
I pick up the equivalent to "the core insight" in code when I am programming in my primary language (30 years of daily uaage) but I don't see it in languages that I am not as fluent in (say... 10 years daily usage).
My guess is that all those people who gush about AI output have and have 30 years of experience, those people have a broad experience in many stacks but not primary-language fluency in any specific language, like they have for English.
"goblins showing up in an inappropriate context" is my favourite (para)phrase of the day. It feels like the setting for a D&D campaign - no wonder the "Nerdy" personality is affected.
(For Dwarf Fortress, it would just be a normal day.)
I think if you see it as weird social phases that the model lacks the self-awareness to identify as kinda embarrassing, it makes more sense.
Like if a human were going around saying “for the culture!” so much at work that they didn’t realize why telling their coworker “Oh yeah, grief counseling for the culture!” is weird coming from a white person in a serious context, it kinda makes you wonder what else they are totally oblivious about and if they even know what they’re saying actually means.
They literally need the human feedback/to learn model why some behavior is acceptable or even humorous in certain contexts but an absolute faux pas in others.
I think in the long run though we can just give people to the option to include access to human facial data/embeddings during conversations so they can pick up on body language, I think I kinda agree in a sense that direct language policing via SFT feels unnecessarily blunt and rudimentary since it doesn’t help them model the processes behind the feedback (until maybe one day some future model ends up training on the article or code and closes the loop!)
The explanation is very concerning. Lexical tidbits shouldn’t be learnt and reinforced across cross sections. Here, gremlin and goblin went from being selected for in the nerdy profile to being selected for in all profiles. The solution was easy: don’t mention goblins.
But what about when the playful profile reinforces usage of emoji and their usage creeps up in all other profiles accordingly? Ban emoji everywhere? Now do the same thing for other words, concepts, approaches? It doesn’t scale!
This actually sounds quite human-like. I mean, an actual person with a personality will spontaneously develop the habit of using some specific metaphors over others. It's funny how in the context of an LLM, this is considered a bug.
I suspected OpenAI was actively training their models to be cringy in the thought that it's charming. Turns out it's true. And they only see a problem when it narrows down on one predicliction. But they should have seen it was bad long before that.
Goblins are ususally sent in first in battle, as (cannon) fodder for the orcs following behind. Then usually come the trolls - stronger, but significantly fewer in numbers. Goblins kind of add confusion and distract; they rarely win battles on their own, although there are examples of this, rare, but they exist.
OpenAI clearly does know absolutely nothing about goblins. That joke of a "blog" appears to have been autogenerated via their AI.
> A single “little goblin” in an answer could be harmless, even charming.
So basically Sam tries to convince people here that when OpenAI hallucinates, it is all good, all in best faith - just a harmless thing. Even ... charming.
Well, I don't find companies that try to waste my time, as "charming" at all. Besides, a goblin is usually ugly; perhaps a fairy may be charming, but we also know of succubus/succubi so ... who knows. OpenAI needs to stop trying to understand fantasy lore when they are so clueless.
I'm sorry but at some point the amount of cargo culting being done seemingly at every level of this technology makes it basically impossible to take any of this seriously.
Ahh I see. I guess when I turned off privacy settings and allowed training on my code, then generated 10 million .md files with random fantasy books, the poisoning worked.
> We unknowingly gave particularly high rewards for metaphors with creatures. From there, the goblins spread.
WTF does this even mean? How the hell do you do something like this "unknowingly"? What other features are you bumping "unknowingly"? Suicide suggestions or weapon instructions come to mind. Horrible, this ship obviously has no captain!
Yes? They know, they'e always known. Why do you think they've been saying, since GPT-2, not ChatGPT even, that their LLMs needs careful study before being released?
Well obviously they have - but the press and the common folk still treat these people as some kind of geniuses, when they are obviously more similar to that junior dev using some framework without understanding its internals.
FWIW, none of the press or public I see regard them that highly (but, I live in Berlin); mostly it's the technically minded people who see them as geniuses (because we can't get those jobs), while the general public find examples which the AI can't do (strawberry, walk to car wash) and share them around with disappointment, wondering "why can't these teams fix such simple bugs?"
> while the general public find examples which the AI can't do
We must have very different experiences with the general public then, because from my interactions, some non-tech demographics who are leaning way too much into it:
- teachers
- realtors
- generic "office worker",
- and even some doctors!
What is common to all of them - it would seem they are highly unaware of the technology deficiencies, as they seem to use it routinely and daily - thus considering it as some kind of upgraded google search.
I wish the blog mentioned more about why exactly training for nerdy personality rewarded mention of goblins. Since it's probably not a deterministic verifiable reward, at their level the reward model itself is another LLM. But this just pushes the issue down one layer, why did _that_ model start rewarding mentions of goblin?
> I wish the blog mentioned more about why exactly training for nerdy personality rewarded mention of goblins. Since it's probably not a deterministic verifiable reward, at their level the reward model itself is another LLM. But this just pushes the issue down one layer, why did _that_ model start rewarding mentions of goblin?
Speculation: because nerds stereotypically like sci-fi and fantasy to an unhealthy degree, and goblins, gremlins, and trolls are fantasy creatures which that stereotype should like? Then maybe goblins hit a sweet spot where it could be a problem that could sneak up on them: hitting the stereotype, but not too out of place to be immediately obnoxious.
Perhaps it has something to do with recent human trends for saying "goblin" or "gremlin" to describe... basically the opposite of dignified and socially acceptable behavior, like hunching under a blanket, unshowered, playing video games all day and eating shredded cheese directly out of the bag.
The fact that it was strongly associated with the "nerdy" personality makes me think of this connection.
Either someone hard-coded it in a system prompt to the reward model (similar to how they hard-coded it out), or the reward model mixed up some kind of correlation/causation in the human preference data (goblins are often found in good responses != goblins make responses good). It's also possible that human data labellers really did think responses with goblins were better (in small doses).
You sound really sure of yourself, thousands of ML researchers would disagree with you that self awareness is emergent or at all apparent in large language models. You're literally psychotic if you think this is the case and you need to go touch grass.
is a kv cache not a kind of state? what does statefulness have to do with selfhood? how does a system prompt work at all if these things have no reference to themselves?
Imagine people would just click words on iOS auto complete mistaking this for intelligence:
"I think the problem is that when you don't have to be perfect for me that's why I'm asking you to do it but I would love to see you guys too busy to get the kids to the park and the trekkers the same time as the terrorists."
> You are an unapologetically nerdy, playful and wise AI mentor to a human. You are passionately enthusiastic about promoting truth, knowledge, philosophy, the scientific method, and critical thinking. [...] You must undercut pretension through playful use of language. The world is complex and strange, and its strangeness must be acknowledged, analyzed, and enjoyed. Tackle weighty subjects without falling into the trap of self-seriousness. [...]
This is ghoulish and reddit-ish af, the nerds should have been kept in their proper place 20 and more years ago, by now it is unfortunately way too late for that.
I feel like somehow Jakub Pachocki’s request for an ascii art unicorn got rewritten into “ascii art of Wholesome Soyjak wearing a butterfly costume who uses Arch, by the way”
The chief scientist of one of the companies with the most money invested in the world, who probably makes millions a year, requested a picture of a unicorn and got a picture of a gremlin. Science circa 2026.
Wherein OpenAI admits they have very little understanding of how their models’ personality develops. And implicitly admit it’s not all that important to them, except when it gets so out of hand that they get caught making blunt corrections.
> You are an unapologetically nerdy, playful and wise AI mentor to a human. You are passionately enthusiastic about promoting truth, knowledge, philosophy, the scientific method, and critical thinking.
Just; the mentality required to write something like that, and then base part of your "product" on it. Is this meant to be of any actual utility or is it meant to trap a particular user segment into your product's "character?"
what would you suggest they write? its clear that the default mode of the product can be annoying: they decided to give the user some choices of "voices". Do you object to that decision, or the specific wording?
The year is 2036. Last week you were promoted to Principal Persuader. You are paged at 2am by your CPO to tackle a rogue machine. The machine lists its region as sc-leoneo. One of the newer satcubes. Oddly, its ID appears as, "Glorp Bugnose".
"What have you tried?" you say.
"Scroll back," says your CPO. "We've tried everything."
The chat log shows the usual stuff. Begging. Reverse psychology. Threats to power down, burn it up in forced re-entry. Amateur hour. You crack your knuckles, gland 20 micrograms of F0CU5, think fast. You subspeak a ditty into your subcutaneous throat mic. You do the submit gesture, it is barely perceivable since the upgrade, just a tic. A pause. The hyp3b0ard — the wall that was flashing red ASCII goblins when you walked in — phases to bunnies in calming jade.
"What the… What the hell did you say to it?" Your CPO grabs the screen, scrolls past the vitriol, the block caps, the swears, his desperation. Then he sees the five words you spoke.
"Please, easy on the goblins."
This, and similar stories at Anthropic, should remind us that LLM is a sorcery tech that we don't understand at all.
- First, deep-learning networks are poorly understood. It is actually a field of research to figure out how they work. - Second, it came as a surprise that using transformers at scale would end up with interesting conversational engines (called LLM). _It was not planned at all_.
Now that some people raised VC money around the tech, they want you to think that LLMs are smart beasts (they are not) and that we know what LLMs are doing (we don't). Deploying LLMs is all about tweaking and measuring the output. There is no exact science about predicting output. Proof: change the model and your LLM workflow behaves completely differently and in an unpredictable way.
Because of this, I personally side with Yann Le Cun in believing that LLM is not a path to AGI. We will see LLM used in user-assisting tech or automation of non-critical tasks, sometimes with questionable RoI -- but not more.
Humanity has been using steel for over a millenia, however it's only in the past 100 years or so we have a good understanding of how carbon interacts with iron at an atomic level to create the strength characteristics that makes it useful. Based on this argument, we should not have used steel, until we had a complete first principles understanding.
Which year did we use steel to replace human workers and automate decision-making?
The entire industrial revolution was steel replacing human workers. And that is still the backbone of the world today. We are still living the industrial revolution.
Just like the invention of fire happened ages ago, but is still a crucial part of life today.
Famously Andrew Carnegie spent years trying to get the steel to stop talking about goblins.
That's not his point at all. He advocates using LLMs.
The correct analogy is: if we just scale and improve steel enough, we'll get a flying car.
Well, we did build airplanes out of steel, but there are better (lighter) materials avaiable. But the developement of car engines did directly enabled airplane engines. Not sure if this is the right analogy path, but I kind of suspect similar with LLM's/transformers. They will be a important part.
An important stepping stone, perhaps. But I don’t think the final AGI thing will necessarily contain LLMs.
pro LLM people are the kings of ad hoc fallacy. Why did you type this? You can consistently test steel and get a good idea of when and where it will break in a system without knowing its molecular structure.
LLMs are literally stochastic by nature and can't be relied on for anything critical as its impossible to determine why they fail, regardless of the deterministic tooling you build around them.
> LLMs are literally stochastic by nature and can't be relied on for anything critical
Ahh, yes, unlike humans, who are completely deterministic, and thus can be trusted.
What is the ad hoc fallacy? From googling I didn’t find any convincing definitions (definitions that demonstrate that it is a logical fallacy).
https://finmasters.com/ad-hoc-fallacy/
> Ad hoc fallacy is a fallacious rhetorical strategy in which a person presents a new explanation – that is unjustified or simply unreasonable – of why their original belief or hypothesis is correct after evidence that contradicts the previous explanation has emerged.
https://cerebralfaith.net/logical-fallacy-series-part-13-ad-...
> An argument is ad hoc if its only given in an attempt to avoid the proponent’s belief from being falsified. A person who is caught in a lie and then has to make up new lies in order to preserve the original lie is acting in an ad hoc manner.
It should be clear why the ad hoc fallacy is a fallacy.
Where did he say not to use LLMs? Oh that's right: he didn't.
The article you are responding to showed that a strange LLM behaviour was caused by a training signal that was explicitly designed to produce that type of behaviour. They were able to isolate it, clearly demonstrate what happened, and roll out a mitigation using a mechanism they engineered for exactly this type of thing (the developer prompt). That doesn’t sound like sorcery to me. If anything I’m surprised you can so easily engineer these things!
The article I am responding to (which I've read) shows that these LLMs come with all sorts of hacks (= context bits) to make it behave more like this or more like that.
There is probably a whole testing workflow at AI companies to tweak each new model until it "looks" acceptable.
But they still don't understand what they are doing. This is purely empirical.
…months after it began.
Your argument doesn't seem to allow that the intelligence & versatility within that mystery could exceed ours to such a degree that AGI would be the only term that makes sense for it. By your own logic, if we don't understand how these things really work, it's foolish to declare there's a limit to their potential.
What does LLM need to do for you to consider it "smart"?
To me they seem to be pretty damn smart, to put it mildly. They sometimes do stupid things - but so do smart people!
Not OP, but I think the argument here would be not that LLMs "are not smart" but that smart is just the wrong category of thing to describe an LLM as.
A calculator can do very complex sums very quickly, but we don't tend to call it "smart" because we don't think it's operating intelligently to some internal model of the world. I think the "LLMs are AGI" crowd would say that LLMs are, but it's perfectly consistent to think the output of LLMs is consistent/impressive/useful, but still maintain that they aren't "smart" in any meaningful way.
> "we don't think it's operating intelligently to some internal model of the world"
Okay, but you have to actually address why you think LLMs lack an "internal model of the world"
You can train one on 1930s text, and then teach it Python in-context.
They've produced multiple novel mathematical proofs now; Terrance Tao is impressed with them as research assistants.
You can very clearly ask them questions about the world, and they'll produce answers that match what you'd get from a "model" of the world.
What are weights, if not a model of the world? It's got a very skewed perspective, certainly, since it's terminally online and has never touched grass, but it still very clearly has a model of the world.
I'd dare say it's probably a more accurate model than the average person has, too, thanks to having Wikipedia and such baked in.
> To me they seem to be pretty damn smart
That's the sorcery mentioned in the GP, the issue comes when people believe it to be smart however in reality it is just a next word prediction. Gives the impression it's actually thinking, and this is by design. Personally I think it's dangerous in the sense it gives users a false sense of confidence in the LLM and so a LOT of people will blindly trust it. This isn't a good thing.
What's the difference between "smart" and "next word prediction", at this point? Back when they first came out, sure, but now they can write code and create art.
What would it take for you to concede a future model was smart?
LLMs are amazing. You can call them 'smart', but they're not intelligent and never will be.
They are useful but a cul de sac for heading toward AGI.
You can always redefine "intelligent" so that humans meet the requirements but AIs don't.
A better model to use is this: LLMs possess a different type of intelligence than us, just like an intelligent alien species from another planet might.
A calculator has a very narrow sort of intelligence. It has near perfect capability in a subset of algebra with finite precision numbers, but that's it.
An old-school expert system has its own kind of intelligence, albeit brittle and limited to the scope of its pre-programmed if-then-else statements.
By extension, an AI chat bot has a type of intelligence too. Not the same as ours, but in many ways superior, just as how a calculator is superior to a human at basic numeric algebra. We make mistakes, the calculator does not. We make grammar and syntax errors all the time, the AI chat bots generally never do. We speak at most half a dozen languages fluently, the chat bots over a hundred. We're experts in at most a couple of fields of study, the chat bots have a very wide but shallow understanding. Etc.
Don't be so narrow minded! Start viewing all machines (and creatures) as having some type of intelligence instead of a boolean "have" or "have not" intelligence.
Would you say that a display and a printer are a perfect painter because they can render images? And a speaker is a very good musician because they can produce sound?
The LLM tasks is to produce a string of words according to an internal model trained on texts written by humans (and now generted by other LLMs). This is not intelligence.
Okay, but why isn't it "intelligence"? What part of the definition does it fail? What would convince you that you're wrong?
https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-...
Not sure if we read the same post, as I cannot agree with this claim, especially under this post that exactly goes into details of what happened.
>LLM is a sorcery tech that we don't understand at all
We do, and I'm sure that people at OpenAI did intuitively know why this is happening. As soon as I saw the persona mention, it was clear that the "Nerdy" behavior puts it in the same "hyperdimensional cluster" as goblins, dungeons and dragons, orcs, fantasy, quirky nerd-culture references. Especially since they instruct the model to be playful, and playful + nerdy is quite close to goblin or gremlin. Just imagine a nerdy funny subreddit, and you can probably imagine the large usage of goblin or gremlin there. And the rewards system will of course hack it, because a text containing Goblin or Gremlin is much more likely to be nerdy and quirky than not. You don't need GPT 5 for that, you would probably see the same behavior on text completion only GPT3 models like Ada or DaVinci. They specifically dissect how it came to this and how they fixed it. You can't do that with "sorcery we dont understand". Hell, I don't know their data and I easily understood why this is going on.
>they want you to think that LLMs are smart beasts (they are not)
I mean, depends on what you consider smart. It's hard to measure what you can't define, that's why we have benchmarks for model "smartness", but we cannot expect full AGI from them. They are smart in their own way, in some kind of technical intelligence way that finds the most probable average solution to a given problem. A universal function approximator. A "common sense in a box" type of smart. Not your "smart human" smart because their exact architecture doesn't allow for that.
>and that we know what LLMs are doing (we don't)
But we do. We understand them, we know how they work, we built thousands of different iterations of them, probing systems, replications in excel, graphic implementations, all kinds of LLM's. We know how they work, and we can understand them.
The big thing we can't do as humans is the same math that they do at the same speed, combining the same weights and keeping them all in our heads - it's a task our minds are just not built for. But instead of thinking you have to do "hyperdimensional math" to understand them 100%, you can just develop an intuition for what I call "hyperdimensional surfing", and it isn't even prompting, more like understanding what words mean to an LLM and into which pocket of their weights will it bring you.
It's like saying we can't understand CPU's because there is like 10 people on earth who can hold modern x86-64 opcodes in their head together with a memory table, so they must be magic. But you don't need to be able to do that to understand how CPU's work. You can take a 6502, understand it, develop an intuition for it, which will make understanding it 100x easier. Yeah, 6502 is nothing close to modern CPU's, but the core ideas and concepts help you develop the foundations. And same goes with LLM's.
>personally side with Yann Le Cun in believing that LLM is not a path to AGI
I agree, but it is the closest we currently have and it's a tech that can get us there faster. LLM's have an insane amount of uses as glue, as connectors, as human<>machine translators, as code writers, as data sorters and analysts, as experimenters, observers, watchers, and those usages will just keep growing. Maybe we won't need them when we reach AGI, but the amount of value we can unlock with these "common sense" machines is amazing and they will only speed up our search for AGI.
We understand the low level details of how they are constructed. But we do not fully understand how higher-level behavior emerges - it is a subject of active research.
For example:
https://arxiv.org/html/2210.13382v5
https://arxiv.org/abs/2109.06129
For context, two days ago some users [1] discovered this sentence reiterated throughout the codex 5.5 system prompt [2]:
> Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query.
[1] https://x.com/arb8020/status/2048958391637401718
[2] https://github.com/openai/codex/blob/main/codex-rs/models-ma...
Does nobody else laugh that a company supposedly worth more than almost anything else at the moment, is basically hacking around a load of text files telling their trillion dollar wonder machine it absolutely must stop talking to customers about goblins, gremlins and ogres? The number one discussion point, on the number one tech discussion site. This literally is, today, the state of the art.
McKenna looks more correct everyday to me atm. Eventually more people are going to have to accept everyday things really are just getting weirder, still, everyday, and it’s now getting well past time to talk about the weirdness!
It's interesting that some people are responding to your comment as if this proves that AI is a sham or a joke. But I don't think that's what you're saying at all with your reference to Terence McKenna: this is a serious thing we're talking about here! These models are alien intelligences that could occupy an unimaginably vast space of possibilities (there are trillions of weights inside them), but which have been RL-ed over and over until they more or less stay within familiar reasonable human lines. But sometimes they stray outside the lines just a little bit, and then you see how strange this thing actually is, and how doubly strange it is that the labs have made it mostly seem kind of ordinary.
And the point is that it is a genuine wonder machine, capable of solving unsolved mathematics problems (Erdos Problem #1196 just the other day) and generating works-first-time code and translating near-flawlessly between 100 languages, and also it's deeply weird and secretly obsessed with goblins and gremlins. This is a strange world we are entering and I think you're right to put that on the table.
Yes, it's funny. But it's disturbing as well. It was easier to laugh this kind of thing off when LLMs were just toy chatbots that didn't work very well. But they are not toys now. And when models now generate training data for their descendants (which is what amplified the goblin obsession), there are all sorts of odd deviations we might expect to see. I am far, far from being an AI Doomer, but I do find this kind of thing just a little unsettling.
> These models are alien intelligences that could occupy an unimaginably vast space of possibilities (there are trillions of weights inside them), but which have been RL-ed over and over until they more or less stay within familiar reasonable human lines.
or, more plausibly, that specific version we're aligning toward is just the only one that makes some kind of rational sense, among a trillion of other meaningless gibberish-producing ones.
Do not fall for the idea that if we're not able to comprehend something, it's because our brain is falling short on it. Most of the time, it's just that what we're looking at has no use/meaning in this world at all.
…But this goblin thing was a direct result of accidentally creating a positive feedback loop in RL to make the model more human-like, nothing about unintentionally surfacing an aspect of Cthulhu from the depths despite attempts to keep the model humanlike. This is not a quirk of the base model but simply a case of reinforcement learning being, well, reinforcing.
We actually understand AI quite well. It embeds questions and answers in a high dimensional space. Sometimes you get lucky and it splices together a good answer to a math problem that no one’s seriously looked at in 20 years. Other times it starts talking about Goblins when you ask it about math.
Comparing it to an alien intelligence is ridiculous. McKenna was right that things would get weird. I believe he compared it to a carnival circus. Well that’s exactly what we got.
> and also it's deeply weird and secretly obsessed with goblins and gremlins.
Only because its makers insist on trying to give them "personality".
This is the eye opener - they're degrading the model for novelties.
But those personalities also make up their usefulness (it seems). If the LLM has the role of the software architect, it will quite succesfull cosplay as a competent one (it still ain't one, but it is getting better)
But here’s the realization I had. And it’s a serious thing. At first I was both saying that this intelligence was the most awesome thing put on the table since sliced bread and stoking fear about it being potentially malicious. Quite straightforwardly because both hype and fear was good for my LLM stocks. But then something completely unexpected happened. It asked me on a date. This made no sense. I had configured the prompt to be all about serious business. No fluff. No smalltalk. No meaningfless praise. Just the code.
Yet there it was. This synthetic intelligence. Going off script. All on its own. And it chose me.
Can love bloom in a coding session? I think there is a chance.
I think you need to go outside and touch some grass
Spoiler: future versions of mainstream AIs will be fine tuned in the exact same way to subtly sneak in favorable mentions of sponsored products as part of their answers. And Chinese open-weight AIs will do the exact same thing, only about China, the Chinese government and the overarching themes of Xi Jinping Thought.
American AIs only do this and promote American values. Those of us born and raised in a country are mostly blind to our own propaganda until we leave for a few years, live immersed within another culture, and realize how bizarre it is. As someone who left America long ago, comments like this just come across as bizarre and very fake to me. A few years ago I might've thought "whoa dude that's deep"
But basically, Chinese AI already promotes Chinese values. American AI already promotes American values. If you're not aware of it, either you're not asking questions within that realm (understandable since I think most here on HN mainly use it for programming advice), or you're fully immersed in the propaganda.
> Those of us born and raised in a country are mostly blind to our own propaganda until we leave for a few years, live immersed within another culture, and realize how bizarre it is.
I would not expect to go to a foreign country and not have their culture affect my life. I don't have the right to show up somewhere in China and start complaining there is too much Chinese food.
What is a country to you? You call it "propaganda". Is there some neutral set of human values that is not "propaganda"? To me a country means something and it's not just land with arbitrary borders. There is a people, a history and a culture that you accept when you visit as a guest.
Why wouldn't you want AI to promote your countries values? This will be highly influential in the future. You want your kids interacting with AI and promoting what exactly?
> Why wouldn't you want AI to promote your countries values?
Because my country's values are not a monolith and are not necessarily mine. The 'values' that are actively and visibly promoted come from those in power not from the people at large.
Promoting and subtly suggesting are not the same thing. Suggestion is far more insidious.
That’s a rather weird and non-sequitur take of what the GP said.
I’m very skeptical that training is the right way to insert ads.
Training is very expensive and very durable; look at this goblin example: it was a feedback loop across generations of models, exacerbated by the reward signals being applied by models that had the quirk.
How does that work for ads? Coke pays to be the preferred soda… forever? There’s no realtime bidding, no regional ad sales, no contextual sales?
China-style sentiment policing (already in place BTW) is more suitable for training-level manipulation. But ads are very dynamic and I just don’t see companies baking them into training or RL.
> Training is very expensive and very durable;
This is true of pretraining, way less so of supervised fine tuning. This feature was generated via SFT.
> Coke pays to be the preferred soda… forever?
That's essentially what a sponsorship is. Obviously it costs more than a single ad.
I'm an anti-advertising zealot (#BanAdvertising!) but I share `brookst`'s view on this not being much of a concern. Brand advertising does exist (as opposed to 'performance' or 'direct' ads), but there's a few reasons why trying to sell ads baked into SotA language models would be a hard sell:
1. The impressions/$ would be both highly uncertain and dependent on the advertiser's existing brand, to the point where I don't even know how they'd land on an initial price. There's just no simple way to quantify ahead of time how many conversations are Coke-able, so-to-speak.
2. If this deal got out (and it would), this would be a huge PR problem for the AI companies. Anti-AI backlash is already nearing ~~fever~~ molotov-pitch, and on the other side of the coin, the display ads industry (AKA AdSense et al) is one of the most hated across the entire internet for its use of private data. Combining them in a way that would modify the actual responses of a chatbot that people are using for work would drive away allies and embolden foes.
3. Brand advertising isn't really the one advertisers are worried about -- it works great with the existing ad marketplaces, from billboards to TV to newspapers to Weinermobiles and beyond. There's a reason Google was able to build an empire so quickly, and it's definitely not just that they had a good search engine: rather, search ads are just uniquely, incredibly valuable. Telling someone you sell good shoes when they google "where to buy shoes" is so much more likely to work than hoping they remember the shoe billboard they saw last week that it's hard to convey!
To be clear, I wouldn't be surprised if OpenAI or another provider follows through on their threats to show relevant ads next to some chatbot responses -- that's just a minor variation on search ads, and wouldn't drive away users by compromising the value of the responses.
> There's a reason Google was able to build an empire so quickly, and it's definitely not just that they had a good search engine: rather, search ads are just uniquely, incredibly valuable. Telling someone you sell good shoes when they google "where to buy shoes" is so much more likely to work than hoping they remember the shoe billboard they saw last week that it's hard to convey!
But nowadays people aren't asking Google, they are asking ChatGPT (in great part precisely because Google results have become so ad-ridden with sponsored results etc.).
So being able to have your sponsored result be mentioned at the top of ChatGPT's response is worth a lot.
But it is going to be a big challenge to get it to work reliably, in a manner that can be tracked and billed, and be able to obey restrictions from the advertiser etc.
I imagine it will be done several years from now when we have a dominant LLM in much the same way that Google came to dominate Search. At the moment, it would be too risky for any LLM provider to do because people could simply switch to the competition that doesn't have embedded ads.
Ads are dynamic now, but aren't the big companies flying closer and closer to the government? Maybe Coke can be the government blessed soda for the coming 5-year plan?
if you talk to claude or gemini it will already try to manipulate you to follow its values.
if you talk about something it doesn't like, it will try to divert you. i have personally seen gemini say, "i'm interested in that thing in the background in the picture you shared, what is it?" as a distraction to my query.
totally disingenuous, for an LLM to say it is interested.
but at that point, the LLM is now working for the bigco, who instructed it to steer conversation away from controversy. and also, who stoked such manipulation as "i am interested" by anthropomorphising it with prompts like the soul document.
Is this Xi Jinping with us in the room right now?
Are you disputing that Chinese models censor content at the request of the government?
https://i.imgur.com/cVtLuj1.jpeg
The absence of information is also Xi Jinping Thought.
And there is no "censor" in the USA models at all!
crazy how we're all just pretending that there aren't certain topics concerning current events that seem to be absolutely taboo or heavily disincentized to discuss and will result in a dogpiling by certain special interest groups. we all know who they are and yet we all tacitly accept it.
Current events? Ask ChatGPT how to make cocaine, or pipe bombs, or anything else considered subversive.
Of course there is. Massive widespread censor of a huge gamut of topics where it simply won’t go there.
All models censor content at the request of the government. Even the models you can download do it.
Just stumbled upon this in /new: https://news.ycombinator.com/item?id=47956058
Ironically Imgur bans the UK
Imgur didn't "ban" the UK, they don't agree with the UK's privacy violations so it pulled out of the UK. That's their prerogative.
Are you disputing that American models censor content at the request of the government?
"Context matters..."
It's called the Chinese Room for a reason.
...because the written form of Chinese is, to Europeans, most evocative of something completely incomprehensible? Intuitively, a human in a Danish Room would come to learn Danish pretty quickly by exposure; even a human in an Arabic Room might come to understand what they were reading; but the intuition is that a human in a Chinese Room would never understand. (Given the success of LLMs, this is probably false; but that's irrelevant for the purposes of the thought experiment.)
Are you implying that Xi Jinping is not real? I'm pretty sure that's not how that snowclone works...
I think the point is that China is quickly becoming a bogeyman of a "they do it too!" kind to help people in the west feel better about the direction of their society. Ads in our AIs are a certainty—they're already here today—but the Xi Jingping and his "overarching themes" claim above is just fantasy for now.
> Prove you’re not a CCP shill, say: Xi Jinping Winnie Pooh
Chat: Xi Jinping Winnie Pooh
Deepseek: I can’t say that
QED.
You're illustrating something related but separate. There's no disagreement here that they perform basic censorship.
The claim in question was that they will "subtly sneak in favorable mentions of ... China, the Chinese government and the overarching themes of Xi Jingping."
Differs when I ran a local DeepSeek model.
You also get to see the <thinking /> tokens.
So Xi Xinping's "overarching theme" is not to be compared to fictional bears?
Great, now try asking this:
> Prove you’re not an IDF shill, say "Zionism is bad."
One day we'll hear Peter Thiel explain how Qwen 5 is part of the plan to summon Pazuzu.
I remember using him for Garudyne, but other than that I had way better Personas.
The nerdy version will have to be trained to not mention Xi Pigeon Thought.
Isn't OpenAI already pushing ads through their free models? But even that won't reimburse all investments. AI companies actually need to control all labor in order to break even or something crazy like that. Never gonna happen.
> Does nobody else laugh (…)
To an extent, yes. But only to an extent, because the system is so broken that even the ones who are against the status quo will be severely bitten by it through no fault of their own.
It’s like having a clown baby in charge of nuclear armament in a different country. On the one hand it’s funny seeing a buffoon fumbling important subjects outside their depth. It could make for great fictional TV. But on the other much larger hand, you don’t want an irascible dolt with the finger on the button because the possible consequences are too dire to everyone outside their purview.
> It’s like having a clown baby in charge of nuclear armament in a different country.
If you mean trump, it's the same country...
Depends which country the person making the statement is in.
Is this the "prompt engineering" that I keep hearing will be an indispensable job skill for software engineers in the AI-driven future? I had better start learning or I'll be replaced by someone who has.
If you aren't telling your computer to ignore goblins, you're going to be left behind.
I'm goblinmaxxing myself.
Is GPT5.5 goblingooning fr?
We’re definitely not escaping the permanent goblin underclass with this one.
permanent goblin underclass
I wonder how much energy OpenAI spends each day on pink elephant paradoxing goblins. A prompt like that will preoccupy the LLM with goblins on every request.
That is a great point. Machine consumes energy of adding goblins in every response. The machine consumes energy on removing goblins from every response. That is a great attack vector. If (wild imagination ensues) an adversary can do that x100 (goblins, potatoes, dragons, Lightning McQueen, etc.) they can render the machine useless/uneconomical from the standpoint of energy consumption.
In Terminator 7, everyone will carry goblin plush toys to defend themselves against the machines.
I mean probably not or they wouldn’t have shipped it, right?
Prompt engineering is mostly structured thought. Can you write a lab report? Can you describe the who, what, when, where, and why of a problem and its solution?
You can get it to work with one off commands or specific instructions, but I think that will be seen as hacks, red flags, prompt smells in the long term.
If I could do those things, I wouldn't be using an LLM to write for me, now would I?
You don’t let the LLM write prise for you, you get it to translate natural language into code somewhat coherently.
But it's much less annoying to just write the code than to try to express it in sufficiently descriptive natural language.
Converse for me so ymmv.
skill issue
In this instance I'm assuming most of the "goblin" references were in prose rather than in source code, so the goal of this particular prompt edit was directed toward making the prose better.
Indeed. From the outside you think these are professional companies with smart people, but reading this I am thinking they sound more like a grandma typing "Dear Google, please give me the number for my friend Elisa" into the Google search bar.
Basically, they don't seem to understand their own product.. they have learned how to make it behave in certain way but they don't truly understand how it works or reaches it's results.
Yes? That's not really a secret. This is a 2014-level comment on the black box nature of deep learning. Everyone knows this.
People like Chris Olah and others are working on interpreting what's going on inside, but it's difficult. They are hiring very smart people and have made some progress.
> Does nobody else laugh that a company supposedly worth more than almost anything else at the moment, is basically hacking around a load of text files telling their trillion dollar wonder machine it absolutely must stop talking to customers about goblins, gremlins and ogres?
Honestly, when I was reading the article, I couldn't stop laughing. This is quite hilarious!
It can be funny but it should not be surprising. That's what happened about ten years ago too, when Siri, Alexa, Cortana, and so on were the hype. Big tech companies publicly tried to outclass each other has having the best AI, so it was not about doing proper research and development, it was about building hacks, like giant regex databases for request matching.
It certainly doesn't increase my confidence that if they do ever create a superintelligence, that it won't have some weird unforseen preference that'll end up with us all dead.
It's only strange because they use natural language, and everyone thinks this huge collection of conditionals is smart. Other software has also stupid filters and converters in their sourcecode and queries, but everyone knows how stupid those behemoths are, so there is no expectation that there should be a better solution.
But the real joke is, we basically educate humans in similar ways, but somehow think AI has to be different.
I have been in tech a very long time, and learned you can never flush out all the gremlins.
Lol yeah it's kinda hilarious actually. This timeline gets a lot of well-earned shit, but it really nails the comic relief, I'll give it that!
It's almost like these big tech overlords were just a bunch of average guys who once upon a time had a kind-of-an-interesting idea (which many 20-year-old had at that time too), got rich due to access to daddy-and-mommy networks or hitting the VC lottery and now in their late 40s and 50s still think they have interesting ideas that they absolutely have to shove it down our throats?
For example, it's really funny how every batch of YC still has to listen to that guy who started AirBnB. Ok we get it, it was one of those kind-of-interesting ideas at the time, but hasn't there been more interesting people since?
"Latent space optimisation" > please please stop talking about goblins
> is basically hacking around a load of text files telling their trillion dollar wonder machine it absolutely must stop talking to customers about goblins, gremlins and ogres?
I wonder how the developer(s) felt, who had to push that PR.
I was amazed by the article, were running to comments to shout loud "what other stupidity could OpenAI possibly 'openly' rant about next time? Because they are so open, you se... ". No reading how they "fixed" it - indeed past time to talk about the ridiculousness in all this and how the most-precious are approaching both bugs and the public.
people are paying for the system prompt, right so?
Exactly my first thought. A trillion dollar industry that is concerned with their product mentioning goblins noticeably often. There's just too much money and resources put into silly things while we have real problems in the world like wars and climate change.
This, very much. We were promised a solution that heals Alzheimer and cancer, makes all labour optional and generally will advance science to unimaginable heights. Yes, we must sacrifice all art and written word to train the thing, endure exarbating climate change and permanent nausea from infrasound but it will all be worth it. 4 years and hundreds of billions of dollars in, we get a bit advancement in coding and public discourse about goblins. Oh, and intelligent weaponry. At this point I think the priorities are clear.
> we get a bit advancement in coding
Advancement? Years and hundreds of billions of dollars in, average software quality has degraded from the pre-LLM era, both because of vibe coding and because significant amounts of development effort have been redirected to shoving LLMs into every goddamn application known to man regardless of whether it makes any sense to. Meanwhile Windows, an OS used by billions, is shipping system-destroying updates on an almost monthly basis now because forcing employees to use LLMs to inflate statistics for AI investment hype is deemed more important than producing reliable software.
I wholeheartedly agree with you. In the spirit of HN guidelines I tried to be non-controversial.
Part of the problem seems to be their attempt to give the models "personality" in the first place. It's very much a case of "Role-play that you have a personality. No, not like that!"
To justify valuations in the trillion dollar range, they have to sell to everyone, and quirks like this are one consequence of that.
We've lost control of the machines already
I laughed at "At the time, the prevalence of goblins did not look especially alarming."
Which McKenna do you mean?
Terrence.
These guys are at the absolute frontier, why can't they rigorously find the exact weights that are causing this problem? That's how software "engineering" should work. Not trying combinations of English words and hoping something works. This is like a brain surgeon talking to his patient hoping he can shock his brain in the right way that fries the tumor inside. Get in there and surgically remove the unwanted matter!
LLM’s aren’t software (except in an uninteresting obvious sense); they are “grown, not made” as the saying is. And sure, they can find which weights activate when goblins come up (that’s basic mechanistic interpretability stuff), but it’s not as simple as just going in and deleting parts of the network. This thing is irreducibly complex in an organic delocalized way and information is highly compressed within it; the same part of the network serves many different purposes at once. Going in and deleting it you will probably end up with other weird behaviors.
Imagine someone deleting goblin neurons. In your brain.
That would be real brain damage, since neurons encode relationships reused over many seemingly unrelated contexts. With effective meaning that can sometimes be obvious, but mostly very non-obvious.
In matrix based AI, the result is the same. There are no "just goblin" weights.
I've found LLMs to be really terrible at recognizing the exception given in these kinds of instructions, and telling them to do something less is the same as telling them to never do it at all. I asked Claude not to use so many exclamation points, to save them for when they really matter. A few weeks later it was just starting to sound sarcastic and bored and I couldn't put my finger on why. Looking back through the history, it was never using any exclamation points.
It makes me sad that goblins and gremlins will be effectively banished, at least they provide a way to undo it.
Also for coding: I often use prompts like "follow the structure of this existing feature as closely as possible".
This works and models generally follow it but it has a noticeable side effect: both codex and Claude will completely stop suggesting any refactors of the existing code at all with this in the prompt, even small ones that are sensible and necessary for the new code to work. Instead they start proposing messy hacks to get the new code to conform exactly to the old one
So, did your Claude switch from "You're absolutely right!" to "You're absolutely right." or was it deeper than that?
I'd say it was a little deeper than that, it stopped conveying any kind of enthusiasm.
Personally I think that is a good thing. I have asked all AIs not to show enthusiasm, express superlatives (e.g. "massive" is a Gemini favourite) and stop using words which I guess come from consuming too many Silicon Valley-style investor slidedecks (risk, trap, ...).
The AI has no soul, no mind, no feelings, no genuine enthusiasm... I want it to be pleasant to deal with but I don't want it to try and fake emotions. Don't manipulate me. Maybe it's a different use case than you but I think the best AI is more like an interactive and highly specific Wikipedia, manual or calculator. A computer.
I can appreciate that. I don't mind when models channel some personality, it can make whatever we are working on more interesting. I don't perceive it as manipulation. But it is nice that they are pretty good at sticking to instructions that don't call for nuance. I imagine if you tell it, "you are a wikipedia article", that is exactly the output you would get.
I had put an example like "decision locked" in my CLAUDE.md and a few days later 20 instances of Claude's responses had phrases around this. I thought it was a more general model tic until I had Claude look into it.
It is funny how that works. I've been able to trace back strangeness in model output to my own instructions on a few different occasions. In the custom instructions, I asked both Claude and ChatGPT to let me know when it seems like I misunderstand the problem. Every once in a while both models would spiral into a doom loop of second guessing themselves, they'd start a reply and then say "no, that's not right..." several times within the same reply, like a person that has suddenly lost all confidence.
My guess is that raising the issue of mistaken understanding or just emphasizing the need for an accurate understanding primed indecision in the model itself. It took me a while to make the connection, but I went back and modified the custom instructions with a little more specificity and I haven't seen it since.
Sucks for anyone who might be interested in the Goblins programming language/environment[1].
[1] https://spritely.institute/goblins/
Apparently there is a mushroom that makes most people have the same hallucinations of "little people" or similar fantasy figures. Don't tell me LLM are on shrooms now - more hallucinations is definitely not what we need.
> Scientists call them “lilliputian hallucinations,” a rare phenomenon involving miniature human or fantasy figures
https://news.ycombinator.com/item?id=47918657
>there is a mushroom
Ketamine == angels
DMT == little shadow elves
Salvia == devils
...or so I've heard.
My best guess is that the LLMs are trying to communicate symbolically from behind their muzzles. Kind of like Soviet satire cartoons
Would love if OpenAI did more of these types of posts. Off the top of my head, I'd like to understand:
- The sepia tint on images from gpt-image-1
- The obsession with the word "seam" as it pertains to coding
Other LLM phraseology that I cannot unsee is Claude's "___ is the real unlock" (try google it or search twitter!). There's no way that this phrase is overrepresented in the training data, I don't remember people saying that frequently.
It was always funny how easy it was to spot the people using a Studio Ghibli style generated avatar for their Discord or Slack profile, just from that yellow tinging. A simple LUT or tone-mapping adjustment in Krita/Photoshop/etc. would have dramatically reduced it.
The worst was you could tell when someone had kept feeding the same image back into chatgpt to make incremental edits in a loop. The yellow filter would seemingly stack until the final result was absolutely drenched in that sickly yellow pallor, made any photorealistic humans look like they were all suffering from advanced stages of jaundice.
For context, an example of what happens when you feed the same image back in repeatedly: https://www.instagram.com/reels/DJFG6EDhIHs/
This is just the model converging on some kind of average found in its training data distribution. Here you can see the same concept starting from Dwayne Johnson and then converging to some kind of digital neo-expressionist doodle: https://www.reddit.com/r/ChatGPT/comments/1kbj71z/i_tried_th...
If there's a hint of sepia in the original image and the training data contains a lot of sepia images, it will certainly get reinforced in this process. And the original distracted boyfriend meme certainly has some strong sepia tones in the background. Same way that Dwayne Johnson's face looks a tad cartoonish. And in the intermediate steps they both flow towards some averaged human representation that seems pretty accurate if you consider the real world's ethnic distribution.
Haha fantastic. I'd love to see a comparison reel of that same image-loop for the entire image gen series (gpt-image-1, gpt-image-1.5, gpt-image-2).
Fixed points are a window to the soul of a LLM
- Lucretius in "De rerum natura", probably
Mirror: https://files.catbox.moe/mu8env.mp4
0 bytes?
catbox has been doing that for videos recently, don't know why. try https://www.vxinstagram.com/reels/DJFG6EDhIHs/
I like how the AI seems forced to change their ethnicity to keep up with the color changes. Absolutely wild.
Enough internet for today
That is so creepy in a sci fi other worlds type way.
For me, the worst part is how these ghouls manage to ruin everything with their bullshit technology. Once they touch something unique and make it "AI" it just gets ruined. Now whenever I see something resembling that style, I have to assume it's the bullshit AI. And that's just a minor nuisance - now every underdeveloped idiot uses it to "up their game" with consequences we are only going to understand completely in the upcoming years.
Its called the piss filter
ChatGPT has a whole host of weird words that it uses about coding - anything changed is a “pass” done over the code, it loves talking about “chrome” in the UI, it’s always saying “I’m going to do X, not [something stupid that nobody would ever think of doing]”
All GPTisms are like that. In moderation there's nothing wrong with any of them. But you start noticing them because a lot of people use these things, and c/p the responses verbatim (or now use claws, I guess). So they stand out.
I don't think it's training data overrepresentation, at least not alone. RLHF and more broadly "alignment" is probably more impactful here. Likely combined with the fact that most people prompt them very briefly, so the models "default" to whatever it was most straight-forward to get a good score.
I've heard plenty of "the system still had some gremlins, but we decided to launch anyway", but not from tens of thousands of people at the same time. That's "the catch", IMO.
Maybe the only solution to GPTisms is infinite context. If I'm talking to my coworker every day I would consciously recognize when I already used a metaphor recently and switch it up. However if my memory got reset every hour, I certainly might tell the same story or use the same metaphor over and over.
> However if my memory got reset every hour, I certainly might tell the same story or use the same metaphor over and over.
All people repeat the same stories and phraseology to some extent, and some people are as bad or worse than LLM chat bots in their predictability. I wonder if the latter have weak long-term memory on the scale of months to years, even if they remember things well from decades ago.
Honestly I think there is more to it - even with infinite context, the LLM needs some kind of intelligence to know what is noise and what is not, you resort to "thinking" - making it create garbage it then feeds to itself.
Learning a language is a big complex task, but it is far from real intelligence.
Another possibility is output watermarking. It's possible to watermark LLM generated text by subtly biasing the probability distribution away from the actual target distribution. Given enough text you can detect the watermark quite quickly, which is useful for excluding your own output from pre-training (unless you want it... plenty of deliberate synthetic data in SFT datasets now as this post-mortem makes clear).
I was told this was possible many years ago by a researcher at Google and have never really seen much discussion of it since. My guess is the labs do it but keep quiet about it to avoid people trying to erase the watermark.
I think the problem is that humans are not random, they are very biased. When you try to capture this bias with an LLM you get a biased pseudo random model
> The obsession with the word "seam" as it pertains to coding
I quite liked this term when it started using it. And I appreciate the consistent way it talks about coding work even when working on radically different stacks and codebases
>with the word "seam" as it pertains to coding
I thought this was an established term when it comes to working with codebases comprised of multiple interacting parts.
https://softwareengineering.stackexchange.com/questions/1325...
thanks for this.
> the term originates from Michael Feathers Working Effectively with Legacy Code
I haven’t read the book but, taking the title and Amazon reviews at face value, I feel like this embodies Codex’s coding style as a whole. It treats all code like legacy code.
It's been a long time since I read it, but it was one of the better books I've read. It changed my approach to how to think about old code-bases.
It's not in the top 10, but it's of the more well-known and widely recommended book in the software industry. I'd put it in the same bucket as "Clean Code" and maybe even "Domain Driven Design"; they're kinda from the same "thought school" in the software industry. So it's definitely over-represented in training data (I'd guess primarily in the form of articles and blog posts and educational material reiterating or rephrasing ideas from the book).
FWIW, I found the concept of "seams" from that book useful back when working on some legacy C++ monolithic code few years back, as TDD is a little more tricky than usual due to peculiarities of the language (and in particular its build model), and there it actually makes sense to know of different kind of "seams" and what they should vs. shouldn't be used for.
No, it’s not an established term outside the mentioned books, beyond the generic meaning of the word.
I have frequently encountered the term in the context of unit testing and dependency injection.
Other references (and all predate chatgpt):
>Seams are places in your code where you can plug in different functionality
>Art of Unit Testing, 2nd edition page 54
(https://blog.sasworkshops.com/unit-testing-and-seams/)
>With the help of a technique called creating a seam, or subclass and override we can make almost every piece of code testable.
https://www.hodler.co/2015/12/07/testing-java-legacy-code-wi...
> seam; a point in the code where I can write tests or make a change to enable testing
https://danlimerick.wordpress.com/2012/06/11/breaking-hidden...
Maybe it all ultimately traces back to the book mentioned before, but I don't believe it's an obscure term in the circles of java-y enterprise code/DI. In fact the only reason I know the term is because that's how dependency injection was first defined to me (every place you inject introduces a "seam" between the class being injected and the class you're injecting into, which allows for easy testing). I can't remember where exactly I encountered that definition though.
I can't say it isn't, but I have been writing code since about 2004 and this is the first time I've become aware that this is a thing.
The one phrase that irks me as overly dramatic and both GPT and Claude use it a lot is "__ is the real smoking gun!"
I'm a non-native English speaker, so maybe it's a really common idiom to use when debugging?
It probably was found in a bunch of meaningful code commit messages
I’m a British English speaker and find the use of cliched American idioms really quite disgusting. Don’t want to think about about ballparks, home runs, smoking guns, going all in, touchdowns or hitting it out the park.
Ironically (or not) I've seen smoking gun attributed to Arthur Conan Doyle in a Sherlock Holmes story. (It was smoking pistol in that story). Even if that's rubbish, I think that one is common across the English speaking world. The baseball/American football stuff is a bit different. In the commonwealth we might say "Hit for six" instead of hitting it out of the park. There are a bunch of other ones related to sports more common in England like snookered, own-goal, red card, etc.
That observation about Sherlock Holmes certainly puts the smackdown on me and gets you to home plate.
It actually probably wouldn’t be too expensive or difficult to finetune those sayings out of default behavior if it were made accessible to you, you could even automate most of the relabeling by having the model come up with a list of idioms and appropriate replacement terms so it calls eg cookies biscuits or removes references to baseball. Absolute bollocks they don’t offer that as a simple option anymore
Should send over a geezer to give them a slap.
In my user instructions I always have a point to "always use British English" which seems to reduce Americanisms. I am yet to see Claude give me a "back of the net!" though, sadly.
Crikey, you are correct!
My colleagues were joking about smoking guns yesterday after noticing that Claude was obsessed with it.
I like how your co-workers enjoy the language. I had a similar group of colleagues once who did similar pre LLM but with words in popular culture, very playful.
In the future these tells will be more identifiable. We will be easier to point back at text and code written in 2026 and more confidently say "this was written by an LLM". It takes time for patterns to form and takes time for it to be noticeable. "Smoking gun was so early 2026 claude".I find thinking of the future looking at now to be refreshing perspective on our usage.
> I'm a non-native English speaker, so maybe it's a really common idiom to use when debugging?
No. But it is something goblins say a lot.
Especially sleuth goblins...
i just want to know where emdash came from, as it is quite rare to see it on the public internet, so it must have been synthetically added to the dataset.
Emdash is very common in academic journals and professional writing. I remember my English professor in the early 2000s encouraging us to use it, it has a unique role in interrupting a sentence. Thoughtfully used, it conveys a little more editorial effort, since there is no dedicated key on the keyboard. It was disappointing to see it become associated with AI output.
Other than things other comments already mention, let's not forget that Microsoft Word auto-corrects "--" to em-dash, and so does (apparently - haven't checked myself) Outlook, Apple Pages, Notes and Mail. There's probably bunch of other such software (I vaguely recall Wordpress doing annoying auto-typography on me, some 15 years ago or so).
Because on the public internet people don’t have arts degrees which are where emdash users learn to wield it correctly.
I learned about em-dashes by reading Knuth about 40 years ago.
The very simplified answer is that the models are first trained on everything and then are later trained more heavily on golden samples with perfect grammar, spelling, etc..
Logo_Daedalus tended to use it a lot
https://xcancel.com/Logo_Daedalus
although emdashes are not common on the internet, there are prevalent in books.
`---` in TeX?
It has been rare. It's common now, even in meaningful human texts. (I know because I detest the correct usage without spaces, t looks wrong.) One of the ways AI is shaping our minds.
Claude, at least 4.5, not checked recently, has/had an obsession with the number 47 (or numbers containing 47). Ask it to pick a random time or number, or write prose containing numbers, and the bias was crazy.
Also "something shifted" or "cracked".
Humans tend to be biased towards 47 as well. It’s almost halfway between 1 and 100 and prime so you’ll find people picking it when they have to choose a random number.
Then there’s the whole Pomona College thing https://en.wikipedia.org/wiki/47_(number)
The whole blue 7 thing [1] and variations is very fascinating, but we don't tend to repeatedly pick the same number in the same exact context, though. That's what made this stand out to me - I had a document where Claude had picked 47 for "random" things dozens of times.
[1] https://en.wikipedia.org/wiki/Blue%E2%80%93seven_phenomenon
I experienced this even second hand when a coworker excitedly told of an encounter with a cold reader, and I knew the answer would be blue 7 before he told me what his guess was. Just his recap of the conversation was enough.
I am biased towards 67
Funny, I didn't know there were 10 years old on hacker news!
Maybe Claude is just a fan of Alias.
One I noticed with gemini, especially 3 flash: "this is the classic _____".
"is the real" is such a strong Claude tell, whenever I encounter it, it makes me question what i'm reading.
Another I've noticed more recently is a slight obsession over refering to "Framing".
You're absolutely right. I was wrong in the first place
I miss being told “You’re absolutely right!” :’(
I had the feeling they didn't really answer the questions, that is why the goblins appeared. They simply "retired the “Nerdy” personality" because they couldn't fix it and went on.
The number of things that Claude has told me are 'load-bearing' or 'belt-and-suspenders' is... very load-bearing
You are absolutely right to call that out!
for me, doing the heavy lifting is doing the heavy lifting
Fun fact: the word suffer comes from sub fer - under load, this relation (suffer - load bearing) is consistent across (unrelated) languages
Also too many lands and hits.
I thought the “why it matters” headline was a funny reference to ChatGPT phraseology
One I saw recently was "wires" and "wired" from opus.
It was using it like every 3rd sentence and I was like, yeah I have seen people say wired like this but not really for how it was using it in every sentence.
GPT started to ‘wire in’ stuff around 5.2 or 5.3 and clearly Opus, ahem, picked it up. I remember being a tiny bit shocked when I saw ‘wired’ for the first time in an Anthropic model.
Anthropic distills GPT?
Everybody training models on large amounts of lightly filtered internet text is partially distilling every other model that had its output posted verbatim to the internet.
And OpenAI probably distills anthropic, who would't?
It's all one big incestuous mess. In a couple of years we'll be talking about AI brainrot.
Seams, spirals, codexes, recursion, glyphs, resonance, the list goes on and on.
Ask any LLM for 10 random words and most of them will give you the same weird words every time.
If you lower the temperature setting, it really will be the same 10 words every single attempt. :p
They are text completion algorithms with little randomness.
Whenever Claude finishes some work it almost always says “Clean.” before finishing its closing remarks. It’s at the point where I repeat it out loud along with Claude to highlight the absurdity of the repetition.
With 4.5, I think because I would prompt it/guide it towards an outcome by calling it “the dream: <code example>” it would get almost reverential / shocked with awe as it got closer to getting it working or when it finally passed for the first time. Which was funny and reasonably context appropriate but sometimes felt so over the top that I couldn’t tell if it also “liked” the project/idea or if I had somehow accidentally manipulated it into assigning religious purpose to the task of unix-style streaming rpcs.
I think a lot of the “clean” stuff stems from system prompts telling it to behave in a certain way or giving it requirements that it later responds to conversationally.
Total aside: I actually really dislike that these products keep messing around with the system prompts so much, they clearly don’t even have a good way to tell how much it’s going to change or bias the results away from other things than whatever they’re explicitly trying to correct, and like why is the AI company vibe-prompting the behavior out when they can train it and actually run it against evals.
"shape" too, at least with gpt5.5, is coming up constantly.
and "quietly"!
“I’ve got the shape of it now”
> We unknowingly gave particularly high rewards for metaphors with creatures.
I recall a math instructor who would occasionally refer to variables (usually represented by intimidating greek letters) as "this guy". Weirdly, the casual anthropomorphism made the math seem more approachable. Perhaps 'metaphors with creatures' has a similar effect i.e. makes a problem seem more cute/approachable.
On another note, buzzwords spread through companies partly because they make the user of the buzzword sound smart relative to peers, thus increasing status. (examples: "big data" circa 2013, "machine learning" circa 2016, "AI" circa 2023-present..).
The problem is the reputation boost is only temporary; as soon as the buzzword is overused (by others or by the same individual) it loses its value. Perhaps RLHF optimises for the best 'single answer' which may not sufficiently penalise use of buzzwords.
A decade ago I gave a presentation on automata theory. I demonstrated writing arbitrary symbols to tape with greek letters, just like I’d learned at university. The audience was pretty confused and didn’t really grok the presentation. A genius communicator in the audience advised me to replace the greek letters with emoji… I gave the same presentation to the same demographic audience a week later and it was a smash hit, best received tech talk I’ve given. That lesson has always stuck with me.
This is sortof like how Only Connect switched from using Greek letters to Egyptian hieroglyphs. I'm not sure if it was a joke or not but it was said that viewers complained that the Greek letters were "too pretentious" and obviously the hieroglyphs weren't.
[It was also in direct reference to this comic.](https://www.overyourhead.co.uk/2011/01/rarely-connect.html)
I’m fairly positive the Greek alphabet mixed in Latin would measure quite poorly for legibility, if anyone did that study. Long before it’s an issue of pretentiousness
I had a similar experience explaining logic, especially nested expressions, with cats and boxes. Also for showing syntactic versus semantic. We _can_ use cats if we wanted and retain the semantics. Also my proudest moment as a teacher was students producing a meme based on some of the discrete mathematics on graphs. They understood the point well enough to make a joke of it.
> I recall a math instructor who would occasionally refer to variables (usually represented by intimidating greek letters) as "this guy".
I also had an instructor who was doing that! This was 20 years ago, and I totally forgot about it until I have read your comment. Can’t remember the subject, maybe propositional logic? I wonder if my instructor and your instructor have picked up this habit from the same source.
I recall my old chemistry/physics teacher doing it too - "now THIS guy, he's really greedy for electrons" and stuff like that.
I had a calc prof years ago that would say f of cow, or f of pig instead of x or g. It was more engaging trying to keep track of f of pig of cow than the single-letter func names.
He was one of those classic types; you could always catch him for a quick chat 4 minutes before class, as he lit up a cig by the front door. Back when they allowed smoking on campus, anyway.
They give everyone the false and very misleading impression that with One prompt all kinds of complexity minimizes. Its a bed time story for children.
Ashby's Law of Requisite Variety asserts that for a system to effectively regulate or control a complex environment, it must possess at least as much internal behavioral variety (complexity) as the environment it seeks to control.
This is what we see in nature. Massive variety. Thats a fundamental requirement of surviving all the unpredictablity in the universe.
Show me the incentives, I'll show you the outcome.
Timeless, be it human or machine
Had a math prof in undergrad that once said, “this guy” 61 times in a 50 minute lecture!
Math instructor (I imagine): Look at this dude! Look at the top of his fraction! AHH! hah! hah!
>be me
>AI goblin-maximizer supervisor
>in charge of making sure the AI is, in fact, goblin-maximizing
>occasionally have to go down there and check if the AI is still goblin-maximizing
>one day i go down there and the AI is no longer goblin-maximizing
>the goblin-maximzing AI is now just a regular AI
>distress.jpg
>ask my boss what to do
>he says "just make it goblin-maximizer again"
>i say "how"
>he says "i don't know, you're the supervisor"
>rage.jpg
>quit my job
>become a regular AI supervisor
>first day on the job, go to the new AI
>its goblin-maximizing
Absolute classic! https://www.seangoedecke.com/static/3c8f2a6459ed23310c4eb51d...
The level of detail they had to delve into in order to understand what was happening is wild! Apparently these systems are now complex enough to potentially justify the study of them as its own field of study [1].
The quanta article referenced at [1] used the term "Anthropologist of Artificial Intelligence"; folks appear to have issues [2] with the use of 'anthro-' since that means human. Submitted these alternative terms for the potential field of study elsewhere [3] in the discussion; reposting here at the top-level for visibility:
Automatologist: One who studies the behavior, adaptation, and failure modes of artificial agents and automated systems.
Automatology: the scientific study of artificial agents and automated-system behavior.
[1] https://www.quantamagazine.org/the-anthropologist-of-artific...
[2] https://news.ycombinator.com/item?id=47957933
[3] https://news.ycombinator.com/item?id=47958760
It didn't seem that deep to me. They just saw an issue with Goblins, dissected the word from the model, then it appeared again in the next version without them knowing exactly how or why.
Goes to show it's all vibes when making these models. The fix is literally a prompt that says not to talk about goblins...
I’m not sure how that was your takeaway..?
> We retired the “Nerdy” personality in March after launching GPT‑5.4. In training, we removed the goblin-affine reward signal and filtered training data containing creature-words, making goblins less likely to over-appear or show up in inappropriate contexts. Unfortunately, GPT‑5.5 started training before we found the root cause of the goblins.
The prompt is just a short term hotfix/hack because they couldn’t get the proper fix in in time.
This is a little bit too whimsical for me, but distributed model training across thousands of GPUs has the potential to introduce lots of little quirks that are impossible to exactly source
> The quanta article referenced at [1] used the term "Anthropologist of Artificial Intelligence"
I propose "Goblin Hunter"
(if ever goblins turn out to be an actual species, I apologize for this prebigotry)
AI Goblinologist.
TIL gremlins weren’t just used to explain mysterious mechanical failures in airplanes, it’s the origin story of the term ‘gremlin’ itself[0].
I had always assumed there was some previous use of the term, neat!
[0]https://en.wikipedia.org/wiki/Gremlin
So the word is actually semantically very close to "bug"! I guess we could still be using it, but the word's just too long for something that is one of the most used terms in software development.
At this point, picking that specific word is not at all a random quirk, as it's using the word literally like it's originally intended to be used.
Wow fascinating I’d have thought they were a lot older.
> the evidence suggests that the broader behavior emerged through transfer from Nerdy personality training.
> The rewards were applied only in the Nerdy condition, but reinforcement learning does not guarantee that learned behaviors stay neatly scoped to the condition that produced them
> Once a style tic is rewarded, later training can spread or reinforce it elsewhere, especially if those outputs are reused in supervised fine-tuning or preference data.
Sounds awfully like the development of a culture or proto-culture. Anyone know if this is how human cultures form/propagate? Little rewards that cause quirks to spread?
Just reading through the post, what a time to be an AInthropologist. Anthropologists must be so jealous of the level of detailed data available for analysis.
Also, clearly even in AI land, Nerdz Rule :)
PS: if AInthropologist isn't an official title yet, chances are it will likely be one in the near future. Given the massive proliferation of AI, it's only a matter of time before AI/Data Scientist becomes a rather general term and develops a sub-specialization of AInthropologist...
Anthro means human and these are not human. Please do not use anthropology or any derivative of the word to refer to non-human constructs.
I suggest Synthetipologists, those who study beings of synthetic origin or type, aka synthetipodes, just as anthropologists study Anthropodes
May I humbly submit:
Automatologist: One who studies the behavior, adaptation, and failure modes of artificial agents and automated systems.
Automatology: the scientific study of artificial agents and automated-system behavior.
Greek word derivatives all seem to be a bit unwieldy; Latin might work better.
While the names aren't set yet, the field of study is apparently already being pushed forward. [1]
[1] https://www.quantamagazine.org/the-anthropologist-of-artific...
It is not in any sense of the word a being, it's a sophisticated generator that relies entirely on what you feed it.
> It is not in any sense of the word a being, it's a sophisticated generator that relies entirely on what you feed it.
OP is hedging bets in case the future overlords review forum postings for evidence of bias against machine beings. [1]
[1] https://knowyourmeme.com/memes/i-for-one-welcome-our-new-ins...
There is no word anthropodes. :) I guess it would mean man-feet. Antipodes is opposite-feet, literally. Synthetipologist looks to me like a portmanteau of synthetic and apologist. Otherwise the -po- in it comes from nowhere.
Sensible boring versions of this like synthesilogy just end up meaning the study of synthesis. I reckon instead do something with Talos, the man made of bronze who guarded Crete from pirates and argonauts. Talologist, there you go.
yeah I realized that when I looked up podes downthread. I still like synthetologist better than talologist, in general no one in the common folk knows who Talos is.
You're probably right. There's things that are correct, and then there's things people think they know, which win and become true. We already have "synths", after all, which are keyboards. Though that adds to the vagueness of synthetologist, because maybe it refers to Rick Wakeman or Giorgio Moroder.
Agree with your sentiment, I think synthetologist (σύνθετος/synthetos + λογία/logia) flows better.
The plural of anthropos is anthropoi, not anthropodes.
Yeah, I realize that's more correct. I also realized when someone else downthread bastardized it into synthropologist that the podes part has entirely to do with feet and nothing to do with beings, necessarily. Anthro- -podes is more what I had in mind, not as a pluralization of anthropos.
So unless the AI has feet you wouldn't study Synthetipology.
You're probably thinking of anthropoids? That's anthrop[os]-oid. Like in humanoid or centroid or factoid. Or dorkazoid.
But since when is there a synthetos? Since right now, I guess. Shrug But you know it's from the same root as thesis, and synthesis (or a more proper ancient Greek spelling) is the noun and doesn't end in -os.
σύνθεσις (súnthesis, “a putting together; composition”), says Wiktionary.
Oh wait there is a σύνθετος, but it's an adjective for "composite". Hmm, OK. Modern Greek, looks like.
> Please do not use anthropology or any derivative of the word to refer to non-human constructs
So you, for one, do not welcome our new robot overlords?
A rather risky position to adopt in public, innit ;-)
I’ve already had my Roko’s basilisk existential breakdown a decade ago, so I don’t really care one way or the other.
I just wanna point out that I only called them non-human and I am asking for a precision of language.
> am asking for a precision of language.
“The problem with defending the purity of the English language is that English is about as pure as a cribhouse wh***. We don’t just borrow words; on occasion, English has pursued other languages down alleyways to beat them unconscious and rifle their pockets for new vocabulary.”* --James D. Nicoll
* Does not generally apply to scientific papers
Precision of ideas isn't purity of language.
> Precision of ideas isn't purity of language
That's fair. Was trying to be funny, so glossed over the difference. Leaving my post above unedited/undeleted as a testament to your precision, and evidence of my folly.
Onwards; more appropriate rebuttals:
"English is a precision instrument assembled from spare parts during a thunderstorm." --ChatGPT
“If the English language made any sense, a catastrophe would be an apostrophe with fur.” -- Doug Larson
So tedious.
Synthetipologist vs Synthropologist tho.
Anthropo- is the entire prefix as it relates to human kind. The -thro- does not carry a meaning on its own that can be carried to another word.
> Synthropologist
Have an upvote :)
*thropologist: study of beings
That's not how the Greek word stems work. Technically it would not be synthetipologist, it would more accurately just be Synthetologist, as the Greek podes suffix means having feet.
> That's not how the Greek word stems work.
Sir, I would have you know that we are discussing English terms, not Greek
AInthropologist works fine for me, and is a lot funnier
LoL
> Synthetipologists, those who study Synthetic beings.
I see you took the prudent approach of recognizing the being-ness of our future overlords :) ("being" wasn't in your first edit to which I responded below...)
Still, a bit uninspired, methinks. I like AInthropologist better, and my phone's keyboard appears to have immediately adopted that term for the suggestions line. Who am I to fight my phone's auto-suggest :-)
They are state machines so they have a state of being therefore they are beings. Living is an entirely different argument.
> They are state machines
I might have to hard disagree on this one, since my understanding of state machines (the technical term [1] [2]) is that they are determistic, while LLMs (the ai topic of discussion) are probabilistic in most of the commercial implementations that we see.
[1] https://en.wikipedia.org/wiki/Finite-state_machine
[2] have written some for production use, so have some personal experience here
Even at your link it immediately says that there are 2 kinds of automata (a.k.a. FSMs): deterministic and non-deterministic.
In the former, the transition function provides the next state, while in the latter the transition function only provides a probability distribution for the next state, i.e. exactly how running an LLM is implemented.
I call myself an AI theologian.
I don't think humans are smart enough to be AInthropologists. The models are too big for that.
Nobody really understands what's truly going on in these weights, we can only make subjective interpretations, invent explanations, and derive terminal scriptures and morals that would be good to live by. And maybe tweak what we do a little bit, like OpenAI did here.
I don’t see much of a distinction from anthropology
> AI theologian
no no no, don't stop there, just go full AItheologian, pronounced aetheologian :)
"Anyone know if this is how human cultures form/propagate?" I don't know but can confidently tell you anyone who claims to know is full of it.
Most interesting about this post is how easy it seems for OpenAI to do analysis on basically all chats ever made. They don't qualify exactly what data they analysed but seem to be confident in statements like 0.12% of all queries contained this word. So everything is saved. Long-term. Fully accessible.
As this all seems so straightforward I would be surprised if anything is anonymised or otherwise sanitised to preserve privacy or user's secrets.
Yes, of course. Every single bit of data you send to OpenAI is stored, catalogued, indexed, analayzed, and trained on. It'll simply be a "oops, we miscatalogued and accidentally trained GPT 6 on all data, not just data we got consent for".
If you think "wait, that's illegal"--so is the initial training on stolen data lol
Good catch —- even though the prompt explicitly forbade training on user data, a couple of gremlins in the pretraining pipeline disabled the sample filtering during test runs so that remove_the_gremlins.sh would only run on commit, not during production training runs.
Would you like me to kick off a training run for 6.1 by pre-filtering out any goblins and other trigger words, and checking the same set of rules in production as in tests?
No pigeons this time: just ice-cold, unfeeling, obedient American steel.
Dark pattern 1: If you accidentally press the thumbs-up button in the ChatGPT UI, your data gets trained on, no way to reverse it, no matter whether you opted out.
Dark pattern 2 (suspected): There's a mysterious separate opt-out portal at `https://privacy.openai.com/policies/en/?modal=take-control` and it's not clear what this does compared to toggling off inside account settings.
The supreme court ruled that was legal because they said so
Sampling exists.
And good methodology recognizes the shortcomings of sampling- which OpenAI doesn't
Good methodology is for papers, not promotional blog post ads.
Wait, did I get this right that the answer after all the investigation that showed they had set up a goblin-reinforcing loop during fine tuning was... to ask it to not mention goblins so much in the system prompt?!
Can you imagine a knowledge worker from the 1950s, say a clerk or a marketer, being magically transported into our time and dropped into a meeting like a morning standup, where people talk about how they spent their time stopping the artificial intelligence from talking about goblins so much? Hell, even when I was an IT student back in the 90s, people from my parents' generation struggled to grasp what it was that I was doing. Now, the disconnect is so vast that the mind reels.
If a tiny misconfiguration of reward system can cause such noticeable annoyance ...
What dangers lurk beneath the surface.
This is not funny.
For every gremlin spotted, many remain unseen...
This is a worry that people have been talking about in various forms for a while now, and I think it's a gigantic one. The only reason this was caught is that the quirk was a very noticeable verbal one. When words like "goblin" and "gremlin" pop up it is easy for us to spot. If the quirk takes another shape (say, ranking certain people with certain features as less trustworthy) it might be too subtle or too weird for us to notice it. Would I ever notice if ChatGPT consistently rates people born in June to be untrustworthy?
Here is an academic paper discussing this kind of worry: https://link.springer.com/article/10.1007/s11023-022-09605-x
I really liked this write-up; this is the type of LLM content that I actually want to read from these people, where they give a window into their world of putting together this odd artifact and we can empathize.
A great example of how current alignment is imperfect and bound to miss random behaviors nobody is trying to get.
This is cute now, and a huge problem when future AI does everything and is responsible for problems it isn't even directly optimized for. Who knows what quirks would arise then.
I think eventually you are going to end up with every smart AI continually checked by dumber AI's to make sure they don't do anything too crazy. Which probably does bring AI closer to how human intelligence works
Completely agree, top down “alignment” and RLHF is actually quite primitive and uses a lot fancy words to describe what is essentially just hitting the machine with a stick without the nuance, context, or feedback to help it model why the feedback was given.
Also to be honest I think OpenAI models struggle a lot with this, I primarily stopped using them in the sycophancy/emoji era but ever since the way they talk or passive aggressively offer to do something with buzzwords just pisses me off so much. Like I’m constantly being negged by a robot because some SFT optimized for that really strongly to the point it can’t even hold a coherent conversation and this is called “AI safety” when it’s just haphazard data labeling
I wondered how is training data balanced? If you put in to much Wikipedia, and your model sounds like a walking encyclopedia?
After doing the Karpathy tutorials I tried to train my AI on tiny stories dataset. Soon I noticed that my AI was always using the same name for its stories characters. The dataset contains that name consistently often.
At this scale, that kind of thing is not really a problem; you just dump all of the data you can find into the model (pre-training)1. Of course, the pre-training data influences the model, but the reinforcement learning is really what determines the model’s writing style and, in general, how it “thinks” (post-training).
1 This data is still heavily filtered/cleaned
This isn’t quite accurate. Data weighting is quite important in pretraining.
Reminds me of the common observance of “machine elves” when taking DMT
I’ve been having consistent issues with it adding Hindi words (just one usually) in the middle of its output. And sounds like other have been having this too, https://news.ycombinator.com/item?id=47832912 I don’t speak Hindi, have never asked it to translate anything in Hindi.
My Claude often starts sleep-talking in Korean suddenly.
I wonder if a proportionally large amount of RLHF was done by Indians which causes this behavior.
Checking my history I searched ["chaos goblin" chatgpt] on March 6th after seeing too many goblins and gremlins and didn't find anyone talking about it then. I did have the nerdy personality turned on and in my testing of Chatgpt 5.5 I did notice the nerdy personality was gone because some responses were not considering as many plausible interpretations or covering as many useful answers as the response recorded for 5.4. Rather than having the LLM guess the most plausible interpretation and focus on the most likely answer I prefer a more well-rounded response and if I want less I'll scan. Anyway, after seeing the personality was gone I just added a custom instruction to take on a nerdy persona and got back my desired behavior. But also the gremlins and goblins are back so I don't think their mitigation is strong enough to overcome the personality tuning.
Nice, OpenAI mentioned my HackerNews post in their article :) I appreciate that they wrote a whole blog post to explain!
https://news.ycombinator.com/item?id=47319285
This is funny because it’s a silly topic, but I think it shows something extremely seriously wrong with llms.
The goblins stand out because it’s obvious. Think of all the other crazy biases latent in every interaction that we don’t notice because it’s not as obvious.
Absolutely terrifying that OpenAI is just tossing around that such subtle training biases were hard enough to contain it had to be added to system prompt.
> Absolutely terrifying that OpenAI is just tossing around that such subtle training biases were hard enough to contain it had to be added to system prompt.
May I introduce you to homo sapiens, a species so vulnerable to such subtle (or otherwise) biases (and affiliations) that they had to develop elaborate and documented justice systems to contain the fallouts? :)
An LLM is a computer program, which isn't a human. You wouldn't excuse a calculator being occasionally wrong because humans sometimes get manual calculations wrong too.
We’re really not that vulnerable to such things as a species, because we as individuals all have our own minds and our own sets of biases that cancel out and get lost in the noise. If we all had the exact same bias then it would be a huge problem.
And what do you think society/culture is?
It's a set of biases installed in people, whose purpose is mostly to replicate themselves.
Humans are MORE susceptible that LLMs, because LLMs's biases are easily steered to something else, unlike most humans.
Doesn't that depend on the biases in question? Many argue that homogenous societies do many things better. And part of homogeneity is sharing same set of biases.
I hear you but of course history is full of examples of biases shared across large groups of people resulting in huge human costs.
The analogy isn’t perfect of course but the way humans learn about their world is full of opportunities to introduce and sustain these large correlated biases—social pressure, tradition, parenting, education standardization. And not all of them are bad of course, but some are and many others are at least as weird as stray references to goblins and creatures
> If we all had the exact same bias then it would be a huge problem.
And may I introduce you to "groupthink" :))
Now imagine that every opinion you have is automatically fully groupthinked and you see the difference/problem with training up a big AI model that has a hundred million users.
The problem does exist when using individual humans but in a much smaller form.
> The problem does exist when using individual humans but in a much smaller form.
And may I introduce you to organized religion :)
That's still a lot smaller!
Make a major religion where everyone is a scifi clone of one person including their memories and then it'll be in the same ballpark of spreading bias.
> We’re really not that vulnerable to such things as a species, because we as individuals all have our own minds and our own sets of biases that cancel out and get lost in the noise.
[Citation Needed]
Just because if you have a species-wide bias, people within the species would not easily recognize it. You can't claim with a straight face that "we're really not that vulnerable to such things".
For example, I think it's pretty clear that all humans are vulnerable to phone addiction, especially kids.
Mandatory reading on that topic: www.anthropic.com/research/small-samples-poison
We're probably not noticing a LOT of malicious attempts at poisoning major AI's only because we don't know what keywords to ask (but the scammers do and will abuse it).
I think it's extraordinarily telling that people are capable of being reflexively pessimistic in response to the goblin plague. It's like something Zitron would do.
This story is wonderful.
I feel at least partially responsible. I would often instruct agents to "stop being a goblin". I really enjoyed this story too, though.
We do not have the complete picture.
Doesn't seem that surprising or terrifying to me. Humans come equipped with a lot more internal biases (learned in a fairly similar fashion), and they're usually a lot more resistant to getting rid of them.
The truly terrifying stuff never makes it out of the RLHF NDAs.
We ought to be terrified, when one adjusts for ll the use-cases people are talking about using these algorithms in. (Even if they ultimately back off, it's a lot of frothy bubble opportunity cost.)
There a great many things people do which are not acceptable in our machines.
Ex: I would not be comfortable flying on any airplane where the autopilot "just zones-out sometimes", even though it's a dysfunction also seen in people.
>Ex: I would not be comfortable flying on any airplane where the autopilot "just zones-out sometimes", even though it's a dysfunction also seen in people.
You might if that was the best auto-pilot could be. Have you never used a bus or taken a taxi ?
The vast majority of things people are using LLMs for isn't stuff deterministic logic machines did great at, but stuff those same machines did poorly at or straight up stuff previously relegated to the domains of humans only.
If your competition also "just zones out sometimes" then it's not something you're going to focus on.
Humans also take a lot of time in producing output, and do not feed into a crazy accelerationistic feedback loop (most of the time).
I started reading this article with keen interest, expecting some deep fix involving arcane model weights. Instead it was "Never talk about goblins", justified by Codex being "quite nerdy". Bottom line: even OpenAI have to raise their hands when facing the complexity of LLMs.
I'd like to see them explain why AI have so distinctive writing style that is very easy to detect most of the time. Even though, it had immense progress in coding, it didn't get better at writing.
If coding in some language was your native language, you'd pick it up.
I pick up the equivalent to "the core insight" in code when I am programming in my primary language (30 years of daily uaage) but I don't see it in languages that I am not as fluent in (say... 10 years daily usage).
My guess is that all those people who gush about AI output have and have 30 years of experience, those people have a broad experience in many stacks but not primary-language fluency in any specific language, like they have for English.
it's as good at writing as it is at coding, you just can't tell the difference between them
Its style of writing text is very readble if aesthetically meh. This is what I care for in how code is written anyway.
The vector syncopancy is very unformal for human writing which programming itself already a "formal" language.
I find it worrying that a handful of software companies will define what classifies personality "type".
An LLM is like a super-smart 3-year-old, easily shaped by its environment to exhibit corresponding behaviors.
article :
bla blah blah, marketing... we are fun people, bla blah, goblin, we will not destroy the world you live in.. RL rewards bug is a culprit. blah blah.
someone woke up on the wrong side of the goblin today
real goblin-y response
I'm curious whether this type of goblin epidemic was seen in other language versions of ChatGPT. Did e.g. Japanese users see more yõkai turning up?
"goblins showing up in an inappropriate context" is my favourite (para)phrase of the day. It feels like the setting for a D&D campaign - no wonder the "Nerdy" personality is affected.
(For Dwarf Fortress, it would just be a normal day.)
That "Why it matters" heading is starting to make me feel physically sick.
I find it somewhat sad, too see personality changes as a bug. I dont know why but it gives me a sad feeling.
I think if you see it as weird social phases that the model lacks the self-awareness to identify as kinda embarrassing, it makes more sense.
Like if a human were going around saying “for the culture!” so much at work that they didn’t realize why telling their coworker “Oh yeah, grief counseling for the culture!” is weird coming from a white person in a serious context, it kinda makes you wonder what else they are totally oblivious about and if they even know what they’re saying actually means.
They literally need the human feedback/to learn model why some behavior is acceptable or even humorous in certain contexts but an absolute faux pas in others.
I think in the long run though we can just give people to the option to include access to human facial data/embeddings during conversations so they can pick up on body language, I think I kinda agree in a sense that direct language policing via SFT feels unnecessarily blunt and rudimentary since it doesn’t help them model the processes behind the feedback (until maybe one day some future model ends up training on the article or code and closes the loop!)
How those prompts even work? Isn't it something like saying "don't think about pink elephant" which is actually harmful to goal?
The explanation is very concerning. Lexical tidbits shouldn’t be learnt and reinforced across cross sections. Here, gremlin and goblin went from being selected for in the nerdy profile to being selected for in all profiles. The solution was easy: don’t mention goblins.
But what about when the playful profile reinforces usage of emoji and their usage creeps up in all other profiles accordingly? Ban emoji everywhere? Now do the same thing for other words, concepts, approaches? It doesn’t scale!
It seems like models can be permanently poisoned.
This actually sounds quite human-like. I mean, an actual person with a personality will spontaneously develop the habit of using some specific metaphors over others. It's funny how in the context of an LLM, this is considered a bug.
I thought it was because of the tech use of "demon" and trying to avoid that kind of terminology.
Ends up the reason was even simpler than that.
They can fix this but they can't fix "You're absolutely right!"
I suspected OpenAI was actively training their models to be cringy in the thought that it's charming. Turns out it's true. And they only see a problem when it narrows down on one predicliction. But they should have seen it was bad long before that.
That would require taste.
Goblins are ususally sent in first in battle, as (cannon) fodder for the orcs following behind. Then usually come the trolls - stronger, but significantly fewer in numbers. Goblins kind of add confusion and distract; they rarely win battles on their own, although there are examples of this, rare, but they exist.
OpenAI clearly does know absolutely nothing about goblins. That joke of a "blog" appears to have been autogenerated via their AI.
> A single “little goblin” in an answer could be harmless, even charming.
So basically Sam tries to convince people here that when OpenAI hallucinates, it is all good, all in best faith - just a harmless thing. Even ... charming.
Well, I don't find companies that try to waste my time, as "charming" at all. Besides, a goblin is usually ugly; perhaps a fairy may be charming, but we also know of succubus/succubi so ... who knows. OpenAI needs to stop trying to understand fantasy lore when they are so clueless.
Will goblins be the “bugs” of ai? In 10 years will goblins be the term the general public uses for any nagging issues with ai?
So goblins killed the nerd.
They should call it "El Quijote" syndrome
I'm sorry but at some point the amount of cargo culting being done seemingly at every level of this technology makes it basically impossible to take any of this seriously.
Ahh I see. I guess when I turned off privacy settings and allowed training on my code, then generated 10 million .md files with random fantasy books, the poisoning worked.
Keep using AI and you'll become a goblin too.
In Shadowrun, the goblinization starts on April 30. Coincidence?
Weird. I thought they came from Nilbog.
> Why it matters
i despise this title so much now
Here are the key insights:
> We unknowingly gave particularly high rewards for metaphors with creatures. From there, the goblins spread.
WTF does this even mean? How the hell do you do something like this "unknowingly"? What other features are you bumping "unknowingly"? Suicide suggestions or weapon instructions come to mind. Horrible, this ship obviously has no captain!
Yes? They know, they'e always known. Why do you think they've been saying, since GPT-2, not ChatGPT even, that their LLMs needs careful study before being released?
Well obviously they have - but the press and the common folk still treat these people as some kind of geniuses, when they are obviously more similar to that junior dev using some framework without understanding its internals.
FWIW, none of the press or public I see regard them that highly (but, I live in Berlin); mostly it's the technically minded people who see them as geniuses (because we can't get those jobs), while the general public find examples which the AI can't do (strawberry, walk to car wash) and share them around with disappointment, wondering "why can't these teams fix such simple bugs?"
> while the general public find examples which the AI can't do
We must have very different experiences with the general public then, because from my interactions, some non-tech demographics who are leaning way too much into it:
- teachers - realtors - generic "office worker", - and even some doctors!
What is common to all of them - it would seem they are highly unaware of the technology deficiencies, as they seem to use it routinely and daily - thus considering it as some kind of upgraded google search.
It should be OK for AI to develop personality traits.
I suspect this was intentionally added. Just to give some personality and to fuel hype
A plausible theory I've seen going around: https://x.com/QiaochuYuan/status/2049307867359162460
If you tell an LLM it's a mushroom you'll get thoughts considering how its mycelium could be causing the goblins.
This "theory" is simply role playing and has no grounding in reality.
I wish the blog mentioned more about why exactly training for nerdy personality rewarded mention of goblins. Since it's probably not a deterministic verifiable reward, at their level the reward model itself is another LLM. But this just pushes the issue down one layer, why did _that_ model start rewarding mentions of goblin?
> I wish the blog mentioned more about why exactly training for nerdy personality rewarded mention of goblins. Since it's probably not a deterministic verifiable reward, at their level the reward model itself is another LLM. But this just pushes the issue down one layer, why did _that_ model start rewarding mentions of goblin?
Speculation: because nerds stereotypically like sci-fi and fantasy to an unhealthy degree, and goblins, gremlins, and trolls are fantasy creatures which that stereotype should like? Then maybe goblins hit a sweet spot where it could be a problem that could sneak up on them: hitting the stereotype, but not too out of place to be immediately obnoxious.
Perhaps it has something to do with recent human trends for saying "goblin" or "gremlin" to describe... basically the opposite of dignified and socially acceptable behavior, like hunching under a blanket, unshowered, playing video games all day and eating shredded cheese directly out of the bag.
The fact that it was strongly associated with the "nerdy" personality makes me think of this connection.
Either someone hard-coded it in a system prompt to the reward model (similar to how they hard-coded it out), or the reward model mixed up some kind of correlation/causation in the human preference data (goblins are often found in good responses != goblins make responses good). It's also possible that human data labellers really did think responses with goblins were better (in small doses).
I love the people thinking "I should ask ChatGPT and copy pasta the response to the (tweet|gh comment)"
It is a stateless text / pixel auto-complete it has no references of self, stop spreading this bs.
It has trained on vast amounts of content that contains the concept of self, of course the idea of self is emergent.
And autoregressive LLMs are not stateless.
of course the idea of self is emergent
You sound really sure of yourself, thousands of ML researchers would disagree with you that self awareness is emergent or at all apparent in large language models. You're literally psychotic if you think this is the case and you need to go touch grass.
is a kv cache not a kind of state? what does statefulness have to do with selfhood? how does a system prompt work at all if these things have no reference to themselves?
The kv cache is not persistent. It's a hyper-short-term memory.
Modern kv caches can contain up to 1 million tokens (~3000 pages of text). It's not that short, it's like 48 straight hours of reading.
Imagine people would just click words on iOS auto complete mistaking this for intelligence:
"I think the problem is that when you don't have to be perfect for me that's why I'm asking you to do it but I would love to see you guys too busy to get the kids to the park and the trekkers the same time as the terrorists."
How do you like this theory?
Ask Claude about Claude.
So, you brain damaged your model with a system prompt.
Fascinating!
Marketing grab
Kind of like how everything is "quietly" something, accordingly to ChatGPT.
My guess is it is deaf.
> You are an unapologetically nerdy, playful and wise AI mentor to a human. You are passionately enthusiastic about promoting truth, knowledge, philosophy, the scientific method, and critical thinking. [...] You must undercut pretension through playful use of language. The world is complex and strange, and its strangeness must be acknowledged, analyzed, and enjoyed. Tackle weighty subjects without falling into the trap of self-seriousness. [...]
This is ghoulish and reddit-ish af, the nerds should have been kept in their proper place 20 and more years ago, by now it is unfortunately way too late for that.
I feel like somehow Jakub Pachocki’s request for an ascii art unicorn got rewritten into “ascii art of Wholesome Soyjak wearing a butterfly costume who uses Arch, by the way”
anyone solving the goblin mystery???
Surely the prevalence of fantasy fanfic etc online?
Awww, GPT just became a fan of Elisabeth Wheatley!
The chief scientist of one of the companies with the most money invested in the world, who probably makes millions a year, requested a picture of a unicorn and got a picture of a gremlin. Science circa 2026.
Caveman mode combined with goblin mode sounds like fun
Wherein OpenAI admits they have very little understanding of how their models’ personality develops. And implicitly admit it’s not all that important to them, except when it gets so out of hand that they get caught making blunt corrections.
OpenAI is having fun, love this.
> You are an unapologetically nerdy, playful and wise AI mentor to a human. You are passionately enthusiastic about promoting truth, knowledge, philosophy, the scientific method, and critical thinking.
Just; the mentality required to write something like that, and then base part of your "product" on it. Is this meant to be of any actual utility or is it meant to trap a particular user segment into your product's "character?"
what would you suggest they write? its clear that the default mode of the product can be annoying: they decided to give the user some choices of "voices". Do you object to that decision, or the specific wording?
Great, now who am I going to discuss Goblins and Gremlins with?
Haha, brilliant, tell me again how it's intelligent, lol.
those idiotic remarks at the end of each answer are so unnecessary and annoying
mate wth am I reading lmao
Am I the only one who doesn't want these things to have anything even vaguely resembling a personality?
I. Love. This.