I am in no way an AI apologist. I’m actually quite critical of the value, investment, products. But I’m also not blind and I’m offended by crap like this.
Claude code is 90% better than previous coding assistants as of weeks ago.
So either your data is invalid or your logic is.
(Such a strong claim only requires one contrary example to disprove, but for everyone at home here are a few more)
Robots can do most manufacturing tasks right now, but can’t move around. Recent advances allow them to move around (but not yet to do the specialized tasks of their fixed location brethren). This advancement is largely due to the flexibility of current model training techniques.
Humanoid robots powered by LLMS + some ten thousand lines of code will be able to move and converse as well as a human and therefore do human level work within 5 years (big stumbling block is literal stumbling which is on a trajectory to solved on the order of months)
LLM + 10,000 lines of code can do most white collar work today. (I now orchestrate 90% of my work with little time commitment) also, did you know that you can orchestrate 10k lines of code in a weekend?
LLM + evaluation of success criteria can solve most repetitive work today.
Even if foundational models have plateaued (They haven’t) the technology percolation will take months and years for the effects to be seen.
If I could be bothered, I’d look up your fund’s bets and bet the other side — I’ve rarely seen someone betting so wrongly. But also I don’t have time to dignify every yahoo out there by betting against them so this note will have to do.
I am in no way an AI apologist. I’m actually quite critical of the value, investment, products. But I’m also not blind and I’m offended by crap like this.
Claude code is 90% better than previous coding assistants as of weeks ago.
So either your data is invalid or your logic is.
(Such a strong claim only requires one contrary example to disprove, but for everyone at home here are a few more)
Robots can do most manufacturing tasks right now, but can’t move around. Recent advances allow them to move around (but not yet to do the specialized tasks of their fixed location brethren). This advancement is largely due to the flexibility of current model training techniques.
Humanoid robots powered by LLMS + some ten thousand lines of code will be able to move and converse as well as a human and therefore do human level work within 5 years (big stumbling block is literal stumbling which is on a trajectory to solved on the order of months)
LLM + 10,000 lines of code can do most white collar work today. (I now orchestrate 90% of my work with little time commitment) also, did you know that you can orchestrate 10k lines of code in a weekend?
LLM + evaluation of success criteria can solve most repetitive work today.
Even if foundational models have plateaued (They haven’t) the technology percolation will take months and years for the effects to be seen.
If I could be bothered, I’d look up your fund’s bets and bet the other side — I’ve rarely seen someone betting so wrongly. But also I don’t have time to dignify every yahoo out there by betting against them so this note will have to do.