Here is the fundamental issue. We use the word "intelligence" for different things. Can you follow a recipe for making sour dough bread? Pretty easy. Can you make sour dough bread? Not so easy. Does following a recipe require "intelligence"? Yes. If something can follow a recipe can it also make bread? Not necessarily.
And another question, perhaps the most important. Can you determine that a recipe is flawed? In immediate terms, if I tell you to feed your sour dough starter every day, can you determine why, how or if that might be bad advice?
My conjecture is that there are at least three types of intelligence, as outlined above. And you have to remember that AI is by definition "artificial". Not in the sense of being unnatural but in the sense of artificial sour dough bread. It is not the real thing. (at least for two out of the three definitions of intelligence).
This is not to argue that AI is not useful and extremely beneficial in some contexts. Unfortunately our whole system of education has trained us to be "follow the recipe" kind of people. Uh Oh! So if your only skill and ability is to follow recipes, you might want to focus on developing your other kinds of intelligence.
I do love the appeal to bread making. It's a wonderful example. If people haven't made french bread by hand, it's a humbling exercise.
Recipes of course have evolved too. Old roman recipes were merely a list of ingredients. Water, flour, salt, yeast.
Written steps came after, then photos, videos, gradually replacing hands on training / kneading.
There are now recipes as code running sour dough assembly lines. Certainly capturing much more detail in technique than even a well made video. But I bet there is still human QA at the end judging "is this bread what folks expect?"
I suspect that in order of complexity you'll get "can I attempt to follow each step", "can I follow the intention of each step and understand if I've failed to meet it" (mitigated by using more specific and detailed steps) "can I follow the intention of the recipe itself - can I add or modify steps that are missing to give the ideal form of sour dough" (maybe you show a machine what good bread looks like, moisture content, crunch?) Those mostly overlap with the 3 you've called out. But I'd add "why would anyone make bread?" Why the heck are we mixing flour and water together to begin with. Why does this recipe exist? Great crusty sourdough requires them all.
In my opinion, the bread example doesn't really work that well because it bridges into the physical domain which most cognitive systems don't have access to. That said, for grounding context and therefore creating truth having a version of a world model is very important (See Yan LeCun's work). My experience is that given the right world to operate in, an agent can indeed find flaws in recipes and fix it even though the agent has not been prompted explicitly to do it. This world, as far as I understand it now, is a combination of sequential (at which step am I in a process), conversational (what was talked about alread/ what had I done already), and context memory (what is the frame or reference/plane of existence).
Reminds me of Moravec's paradox, that it is easy to get computers to ace complex math tests but difficult to teach them to walk. We are very excited that computers have mastered the "know the recipe" step and are underplaying the complexity of actual intelligence required to really replace people.
My fear in your above example would be that we offload more and more of the "know the recipe" intelligence to computers and humans are slotted in as replaceable manual labor and are left arguing with a computer about whether the starter needs to be fed or not (or whatever equivalent scenario).
In the section "Everyone should learn some coding":
> I would say that the major unlocks are at:
1-2 weeks: Basic understanding of what the field is about and what general words to use when asking the AI to do something.
1-2 months: Basic understanding of how and when to ask the AI something.
4-6 months: Ability to check the output for correctness (using external sources as needed).
1-2 weeks for enough of an understanding to appropriately use terms? No way. Using Harvard CS50 as a reference, it takes until week 2 to learn about arrays.
4-6 months to check output for correctness? Are we trusting fresh bootcampers in their first week at their first job to do prod code reviews now?
You can learn a LOT in a short period of time, but it would take much more than casual time investment. This is insane advice on the level of telling blue collar workers to just "learn to code."
Sorry for the off-topic comment, but what happened to the front page? At the time I’m writing this, 11/30 submissions are related to AI. Maybe my comment is cliché too, but I’m honestly tired of all the AI stuff.
Everyone is quite worried of their job. Many of us have made coding/IT our personality, what we were proud of, what the society made us feel valued. It's a big change in life... and there is no solution yet.
So everyone feels the needs to talk about it, to either get rid of this anxiety by ranting or trying to prove that it would be an opportunity, or a non-event depending on the point of view, etc
The crowd who came for the love of computers either left or switched to be the ones who were fed up with building things out of thin air, and are fine with swiping their credit cards to have someone else do the work for them while they're micromanaging that someone else (AI).
Very nice HN client and he was responsive to ideas. I was thinking of same to filter out "Democrat" "Republican" "Trump" and "Musk", partly due to upcoming elections in November.
Welcome to the next wave blogspam campaign for LLMs. Plenty of popular HN blogs have gotten good notoriety by writing about LLMs even if the content is a nothingburger. Now everyone is jumping on that trend to try to continue to normalize it.
Part 1 was flood with AI content. Now Part 2 is walk back bold claims made in Part 1 (call it a fast moving landscape) and have the evangelists flood with AI content. Extra points if you can wax on something and try to redefine it as a pro for LLMs. “What is expertise? Did you have it before? Well now it’s faster with LLMs! Forget about all the efficiency claims, expertise is the real benefit you get with statistically incorrect LLMs.”
Actually - to disillusion yourself from AI, try dabbling into something you do not know. Try writing a production quality 3D engine. Trust me, a 3D engine has its own domain knowledge besides just graphics. No, seriously. And then see how helpless you feel when you yourself do not have the expertise to judge whether the direction being taken is the right or wrong.
At that time, you wish if there were some pipe through which you could reach John Carmack, Tim Sweeney, Gabe Nawell, Jonathan Blow some Casey Muratori and just ask one thing:
Sir, is this really the right direction?
These tools feel good when you yourself are a domain expert. I have written backend systems and designed REST APIs all my life in multiple languages in Java, Python, Go, Ruby for multiple verticals I'd say I am damn expert at API design including all the layers that go under it and I can confidently give a shut up call to an LLM knowing what I know.
Fuck the bean counters and the greedy parasite execs and VPs. Hug a junior today, society will need them tomorrow because I was a clueless junior once and my seniors were very kind to me that I am able to put bread for my family on the table.
Yeah ok. First of all, just because it sucks now, it doesn't mean we're still safe in just 24 months. Everyone was mocking AI 24 months ago.
Second, most of the work out there is not at all about "production quality 3D engine," that's the whole point. Most of us have been doing the same repetitive work for decades. Move this button here. Fix the bug here.
Sure it's not as easy as it looks, but if the average guy can spit out an acceptable app/page in 60 seconds, most people won't even be able to tell the difference.
I think that the universities have an opportunity here to be the places where manual code is written so that juniors can gain the coding expertise necessary to become effective with AI.
Many universities are not set up to take advantage of this opportunity because they lean heavily into theory and look down on coding, but some departments will make the pivot well. I hope that ours (Montana State) is one of them.
The argument for universities to be a place to learn to think critically and not learn specific skills is an even stronger value prop in an era where useful skills likely change rapidly.
so universities become trade schools? one concern is where does one get theoretical knowledge required for e.g. going to graduate school and then doing research to push the state of the art. that's one of the reasons universities emphasize theory: it's seen as the first step on the academic ladder, not as a trade school
LLM are quite a good learning opportunity, mostly in classes where learning is sequential/needs building blocks, like mathematics, where if you miss a trimester, it's finished. Here it's like a free and immediately accessible private tutor. It would be great for computer sciences classes indeed.
> Currently, the level of computing intuition needed to additively prompt the coding agents sits at roughly 5 years’ experience level. Today’s seniors were lucky enough to get paid to build their computing intuition, but the gap grows as coding agents continue to improve.
I am struggling to interpret what they mean by "gap". Gap between what two things?
The gap between juniors and seniors?
The gap between ${AI + 0 YOE} and ${AI + N YOE}? Where N is the growing "gap"? Eg, as AI gets better, you need more and more YOE to justify throwing a salaried human into the loop?
I don't see this discussed often enough but high school and universities need to adapt FAST, like yesterday, to the current reality.
More in-class study and "hands-on" work with proctored in-person exams. There is no incentive for students to go through their courses "the honest way" and build this intuition themselves. Can you blame them?
> More in-class study and "hands-on" work with proctored in-person exams.
If you move to in-class, hands-on work you don't need exams at all as you will see their performance develop in class as well. Exams are for things you can't see them actually use first hand.
I studied computer science and mathematics, not software engineering
Could've used a better software engineering class but I use the more abstract knowledge regularly and I think it would be a disadvantage to strip that out and just go straight to "here's how to prompt"
Sorry if I'm straw manning your comment, I do think that the abstract stuff is more important than ever, and would also like to see more philosophy and such required for eng/science/math degrees.
Schools have always lagged and can barely keep up. Books once printed on any tech topic is almost always outdated by the time it reaches students. Anecdotally, I went through high school being told over and over that I wouldn't always have a calculator in my pocket. I think the messaging they conveyed was done poorly, and should have said "you need to understand the fundamentals and why the calculator gave you the answer".
I think 'expertise' is a bit of a red herring when what is being discussed is experience.
I've always believed that coding and development is an art and something analogous is the experience of a visual arts student. There's a level of experience required when one applies to an art school. The student builds a portfolio of passion projects and demonstrates a passion and skill along with creativity and other beneficial traits. If they are accepted, they learn the deeper theory, techniques, and more that will aide them in their career. This increases their exposure and overall experience.
Experience for a young developer is going to start with passion projects and be supplemented and bolstered through education in a similar way. You can take shortcuts as an arts student or a developer but you really just end up hurting yourself.
What’s built with all that VC money is already built though; I don’t foresee a future a few years out where we don’t have access to an open-source model roughly as good as the current flagship models for the cost of the compute itself.
I have the theory (not tested, subjective) that current economy prefers buying capital (broadly here defined as machine/tools) than having to pay workers salaries, even if both have the same level of competitivity
Capital expenditures are easy to calculate, and it's easy to help raising money. As the current economical system is based on debts, it works quite well: if a company knows that productivity output will raise by 15% over the next year if they spend X dollars, it's easy to get investments (investments firms themselves are relying heavily on private credits, which more and more is coming from bank too). With a system based on debts, they care less about the amount spent, than the yield generated.
With investing in people, it's harder to predict.
Industry does it by buying machines, now knowledge-based companies might do it with GPUs or tokens.
I get the analogy of the calculator. The thing however, is that in college, we had dedicated time to learn how to not use it: classes without it, exams without it, etc.
In current job market and pressure, we doesn't have time anymore. You need to be constantly delivering the new jira ticket, and the time expected to perform a task now decreased, as it's expected of the workers that now they are "more productive with AI".
Nice to see that HN is coming to its senses and people are realizing the flaws and BS in AI / LLMs. We are past-peak Bitcoin / NFT on the curve and I can't wait for this wave to end and move to the next thing.
I don't understand why so many people think that true expertise would become less valuable in the age of AI. How would a non-technical person, who doesn't know the difference between HTTP and HTTPS, have what it takes to build anything serious? I mean, how would you even know to ask the AI for everything that your system needs to be doing, without understanding the concepts?
> And yet, OpenAI, Anthropic, and many top companies continue to compete fiercely for junior talent.
Are they? I would imagine they have the luxury to pick the brightest candidates, and set them to work on jobs for which their models don't have training data for, such as developing new models. Not writing React code.
This is already studied, people do not retain knowledge when learning with AI. Learning with AI only creates the most mediocre of people, I've witnessed this myself over and over and over again over the last couple of years.
Read a book, write, think and you'll be fine. Use LLM and your brain is going to become completely reliant on its ability to access some billionaires thinking machine in order to read and write. You will be a second class citizen who has no differentiating skills. You will end up not being able to write anything on your own or solve problems independently without paying a billionaire, just like how nobody can navigate without Google Maps anymore.
Here is the fundamental issue. We use the word "intelligence" for different things. Can you follow a recipe for making sour dough bread? Pretty easy. Can you make sour dough bread? Not so easy. Does following a recipe require "intelligence"? Yes. If something can follow a recipe can it also make bread? Not necessarily.
And another question, perhaps the most important. Can you determine that a recipe is flawed? In immediate terms, if I tell you to feed your sour dough starter every day, can you determine why, how or if that might be bad advice?
My conjecture is that there are at least three types of intelligence, as outlined above. And you have to remember that AI is by definition "artificial". Not in the sense of being unnatural but in the sense of artificial sour dough bread. It is not the real thing. (at least for two out of the three definitions of intelligence).
This is not to argue that AI is not useful and extremely beneficial in some contexts. Unfortunately our whole system of education has trained us to be "follow the recipe" kind of people. Uh Oh! So if your only skill and ability is to follow recipes, you might want to focus on developing your other kinds of intelligence.
I do love the appeal to bread making. It's a wonderful example. If people haven't made french bread by hand, it's a humbling exercise.
Recipes of course have evolved too. Old roman recipes were merely a list of ingredients. Water, flour, salt, yeast.
Written steps came after, then photos, videos, gradually replacing hands on training / kneading.
There are now recipes as code running sour dough assembly lines. Certainly capturing much more detail in technique than even a well made video. But I bet there is still human QA at the end judging "is this bread what folks expect?"
I suspect that in order of complexity you'll get "can I attempt to follow each step", "can I follow the intention of each step and understand if I've failed to meet it" (mitigated by using more specific and detailed steps) "can I follow the intention of the recipe itself - can I add or modify steps that are missing to give the ideal form of sour dough" (maybe you show a machine what good bread looks like, moisture content, crunch?) Those mostly overlap with the 3 you've called out. But I'd add "why would anyone make bread?" Why the heck are we mixing flour and water together to begin with. Why does this recipe exist? Great crusty sourdough requires them all.
In my opinion, the bread example doesn't really work that well because it bridges into the physical domain which most cognitive systems don't have access to. That said, for grounding context and therefore creating truth having a version of a world model is very important (See Yan LeCun's work). My experience is that given the right world to operate in, an agent can indeed find flaws in recipes and fix it even though the agent has not been prompted explicitly to do it. This world, as far as I understand it now, is a combination of sequential (at which step am I in a process), conversational (what was talked about alread/ what had I done already), and context memory (what is the frame or reference/plane of existence).
Self-correcting agents are already here: https://jdsemrau.substack.com/p/hyperagents-and-self-correct...
Reminds me of Moravec's paradox, that it is easy to get computers to ace complex math tests but difficult to teach them to walk. We are very excited that computers have mastered the "know the recipe" step and are underplaying the complexity of actual intelligence required to really replace people.
My fear in your above example would be that we offload more and more of the "know the recipe" intelligence to computers and humans are slotted in as replaceable manual labor and are left arguing with a computer about whether the starter needs to be fed or not (or whatever equivalent scenario).
In the section "Everyone should learn some coding":
> I would say that the major unlocks are at:
1-2 weeks for enough of an understanding to appropriately use terms? No way. Using Harvard CS50 as a reference, it takes until week 2 to learn about arrays.4-6 months to check output for correctness? Are we trusting fresh bootcampers in their first week at their first job to do prod code reviews now?
You can learn a LOT in a short period of time, but it would take much more than casual time investment. This is insane advice on the level of telling blue collar workers to just "learn to code."
This reminds me of a "rich people meme". You know, people who are rich enough they never have to look at price tags?
"How much could a bunch of bananas cost. $30?"
This strikes me as someone who has lost touch with how much time and effort that building real expertise takes.
Sorry for the off-topic comment, but what happened to the front page? At the time I’m writing this, 11/30 submissions are related to AI. Maybe my comment is cliché too, but I’m honestly tired of all the AI stuff.
Everyone is quite worried of their job. Many of us have made coding/IT our personality, what we were proud of, what the society made us feel valued. It's a big change in life... and there is no solution yet.
So everyone feels the needs to talk about it, to either get rid of this anxiety by ranting or trying to prove that it would be an opportunity, or a non-event depending on the point of view, etc
I don't think you can ignore it. It's the biggest change to tech in 30 years I'd say.
"I'm tired of all this internet talk" in 1990s?
The crowd who came for the love of computers either left or switched to be the ones who were fed up with building things out of thin air, and are fine with swiping their credit cards to have someone else do the work for them while they're micromanaging that someone else (AI).
Even worse it that it's the same few talking points repeated over, and over, and over again - re-spun with AI
Been away for a while? It's been like this for at least a year.
AI is the most important, impactful and disruptive technology of today. It makes sense to be talking about it.
because this is a billboard for YC companies and YC cultural psyops, this isn't an organic forum
5 years ago it was web3. 5 years before that it was Haskell and monads. It’s how things work
Personalization would solve that.
Maybe hit this fellow up with a feature request to add a regex filter, etc: https://github.com/IronsideXXVI/Hacker-News
Very nice HN client and he was responsive to ideas. I was thinking of same to filter out "Democrat" "Republican" "Trump" and "Musk", partly due to upcoming elections in November.
Honestly, at this point it's an accurate reflection of the reality of tech.
It's polluting the world, gobbling up hardware, and making us dumber. And HN and LinkedIn just can't get enough.
People say AI is the new internet. I say AI is the new tobacco.
Welcome to the next wave blogspam campaign for LLMs. Plenty of popular HN blogs have gotten good notoriety by writing about LLMs even if the content is a nothingburger. Now everyone is jumping on that trend to try to continue to normalize it.
Part 1 was flood with AI content. Now Part 2 is walk back bold claims made in Part 1 (call it a fast moving landscape) and have the evangelists flood with AI content. Extra points if you can wax on something and try to redefine it as a pro for LLMs. “What is expertise? Did you have it before? Well now it’s faster with LLMs! Forget about all the efficiency claims, expertise is the real benefit you get with statistically incorrect LLMs.”
Actually - to disillusion yourself from AI, try dabbling into something you do not know. Try writing a production quality 3D engine. Trust me, a 3D engine has its own domain knowledge besides just graphics. No, seriously. And then see how helpless you feel when you yourself do not have the expertise to judge whether the direction being taken is the right or wrong.
At that time, you wish if there were some pipe through which you could reach John Carmack, Tim Sweeney, Gabe Nawell, Jonathan Blow some Casey Muratori and just ask one thing:
Sir, is this really the right direction?
These tools feel good when you yourself are a domain expert. I have written backend systems and designed REST APIs all my life in multiple languages in Java, Python, Go, Ruby for multiple verticals I'd say I am damn expert at API design including all the layers that go under it and I can confidently give a shut up call to an LLM knowing what I know.
Fuck the bean counters and the greedy parasite execs and VPs. Hug a junior today, society will need them tomorrow because I was a clueless junior once and my seniors were very kind to me that I am able to put bread for my family on the table.
Yeah ok. First of all, just because it sucks now, it doesn't mean we're still safe in just 24 months. Everyone was mocking AI 24 months ago.
Second, most of the work out there is not at all about "production quality 3D engine," that's the whole point. Most of us have been doing the same repetitive work for decades. Move this button here. Fix the bug here.
Sure it's not as easy as it looks, but if the average guy can spit out an acceptable app/page in 60 seconds, most people won't even be able to tell the difference.
> Try writing a production quality 3D engine.
Actually I tried that and you are correct about this.
With Claude it took me hundreds of iterations and I'm still not happy.
I tried it with embedded programming, and failed miserably.
I think that the universities have an opportunity here to be the places where manual code is written so that juniors can gain the coding expertise necessary to become effective with AI.
Many universities are not set up to take advantage of this opportunity because they lean heavily into theory and look down on coding, but some departments will make the pivot well. I hope that ours (Montana State) is one of them.
The argument for universities to be a place to learn to think critically and not learn specific skills is an even stronger value prop in an era where useful skills likely change rapidly.
so universities become trade schools? one concern is where does one get theoretical knowledge required for e.g. going to graduate school and then doing research to push the state of the art. that's one of the reasons universities emphasize theory: it's seen as the first step on the academic ladder, not as a trade school
LLM are quite a good learning opportunity, mostly in classes where learning is sequential/needs building blocks, like mathematics, where if you miss a trimester, it's finished. Here it's like a free and immediately accessible private tutor. It would be great for computer sciences classes indeed.
They don't know it yet but universities have a role to curate training data, so we can have trustable models.
Agreed, but I can immediately see how painful it will be to monitor whether the work is actually done by the student.
> Currently, the level of computing intuition needed to additively prompt the coding agents sits at roughly 5 years’ experience level. Today’s seniors were lucky enough to get paid to build their computing intuition, but the gap grows as coding agents continue to improve.
I am struggling to interpret what they mean by "gap". Gap between what two things?
The gap between juniors and seniors?
The gap between ${AI + 0 YOE} and ${AI + N YOE}? Where N is the growing "gap"? Eg, as AI gets better, you need more and more YOE to justify throwing a salaried human into the loop?
I don't see this discussed often enough but high school and universities need to adapt FAST, like yesterday, to the current reality.
More in-class study and "hands-on" work with proctored in-person exams. There is no incentive for students to go through their courses "the honest way" and build this intuition themselves. Can you blame them?
> More in-class study and "hands-on" work with proctored in-person exams.
If you move to in-class, hands-on work you don't need exams at all as you will see their performance develop in class as well. Exams are for things you can't see them actually use first hand.
I studied computer science and mathematics, not software engineering
Could've used a better software engineering class but I use the more abstract knowledge regularly and I think it would be a disadvantage to strip that out and just go straight to "here's how to prompt"
Sorry if I'm straw manning your comment, I do think that the abstract stuff is more important than ever, and would also like to see more philosophy and such required for eng/science/math degrees.
Schools have always lagged and can barely keep up. Books once printed on any tech topic is almost always outdated by the time it reaches students. Anecdotally, I went through high school being told over and over that I wouldn't always have a calculator in my pocket. I think the messaging they conveyed was done poorly, and should have said "you need to understand the fundamentals and why the calculator gave you the answer".
Are proctored in-person exams not the default for most places anyway?
I think 'expertise' is a bit of a red herring when what is being discussed is experience.
I've always believed that coding and development is an art and something analogous is the experience of a visual arts student. There's a level of experience required when one applies to an art school. The student builds a portfolio of passion projects and demonstrates a passion and skill along with creativity and other beneficial traits. If they are accepted, they learn the deeper theory, techniques, and more that will aide them in their career. This increases their exposure and overall experience.
Experience for a young developer is going to start with passion projects and be supplemented and bolstered through education in a similar way. You can take shortcuts as an arts student or a developer but you really just end up hurting yourself.
AI is cheap right now. Let's re-ask this question when it's priced to recover profit and ROI.
What’s built with all that VC money is already built though; I don’t foresee a future a few years out where we don’t have access to an open-source model roughly as good as the current flagship models for the cost of the compute itself.
The hardware will improve in a big way, a lot of money is going into that direction. Llm costs will go down significantly.
I have the theory (not tested, subjective) that current economy prefers buying capital (broadly here defined as machine/tools) than having to pay workers salaries, even if both have the same level of competitivity
Capital expenditures are easy to calculate, and it's easy to help raising money. As the current economical system is based on debts, it works quite well: if a company knows that productivity output will raise by 15% over the next year if they spend X dollars, it's easy to get investments (investments firms themselves are relying heavily on private credits, which more and more is coming from bank too). With a system based on debts, they care less about the amount spent, than the yield generated.
With investing in people, it's harder to predict.
Industry does it by buying machines, now knowledge-based companies might do it with GPUs or tokens.
I get the analogy of the calculator. The thing however, is that in college, we had dedicated time to learn how to not use it: classes without it, exams without it, etc.
In current job market and pressure, we doesn't have time anymore. You need to be constantly delivering the new jira ticket, and the time expected to perform a task now decreased, as it's expected of the workers that now they are "more productive with AI".
Nice to see that HN is coming to its senses and people are realizing the flaws and BS in AI / LLMs. We are past-peak Bitcoin / NFT on the curve and I can't wait for this wave to end and move to the next thing.
hiring top junior talent is more competitive than it's ever been!
I don't understand why so many people think that true expertise would become less valuable in the age of AI. How would a non-technical person, who doesn't know the difference between HTTP and HTTPS, have what it takes to build anything serious? I mean, how would you even know to ask the AI for everything that your system needs to be doing, without understanding the concepts?
> And yet, OpenAI, Anthropic, and many top companies continue to compete fiercely for junior talent.
Are they? I would imagine they have the luxury to pick the brightest candidates, and set them to work on jobs for which their models don't have training data for, such as developing new models. Not writing React code.
AI compresses the time to acquire expertise.
A high schooler can become an expert very quickly with AI, that used to require years and years of education and experience.
but the real expertise still will be to translate real world problems to technical solutions and iterate on design.
This is already studied, people do not retain knowledge when learning with AI. Learning with AI only creates the most mediocre of people, I've witnessed this myself over and over and over again over the last couple of years.
Read a book, write, think and you'll be fine. Use LLM and your brain is going to become completely reliant on its ability to access some billionaires thinking machine in order to read and write. You will be a second class citizen who has no differentiating skills. You will end up not being able to write anything on your own or solve problems independently without paying a billionaire, just like how nobody can navigate without Google Maps anymore.
https://arxiv.org/abs/2506.08872
not really sure how you're imagining AI sidesteps education and experience