Live data from Hacker News

Ask HN: Have LLMs Plateaued?

news.ycombinator.com

31–35 of 35 posts

Re: Ask HN: Have LLMs Plateaued?

#31

Earlier quoted context omitted.

The advances in robotics makes it look like a humanoid robot that could harvest lettuce is only a decade or two away. Making the robot is the easy part. The code to drive it has been the hard part. Until now, that is.

California has about 800,000 agricultural laborers and about 400,000 FTE’s. That’s a lot of robots…and they probably won’t be able to drive themselves from field to field.

Self driving cars are here today, driving in city traffic. Self driving a truck from one field to another seems like it would be eminently doable as far as the technology goes. However, I doubt a bunch of humanoid robots is the optimal robot configuration for harvesting lettuce. Tesla made 1.6 million cars in 2025. Toyota made 11 million. If we handwave that a humanoid robot is roughly as hard to manufacture as a car at scale, there's gonna be a lot of robots.

Re: Ask HN: Have LLMs Plateaued?

#33
I don't think so. In a recent study I did, it was brought up that we are soon or have hit a point where humans are the the bottleneck. So either slow down the rate of progression or figure out a way for humans to retain more information and/or not suck so much.

I mean this in the sense of the human user of the LLM, not the human builders of the LLM; but I guess it is both really.

Re: Ask HN: Have LLMs Plateaued?

#34

I think the giant leap in AI technology was the Transformer. ChatGPT is based on it, with an incredible chat user interface. In terms of technology, there has been no similarly significant innovation since then. But user interfaces and training data keep improving, which continues to increase our productivity.

> with an incredible chat user interface

what makes it incredible?

Re: Ask HN: Have LLMs Plateaued?

#35

Earlier quoted context omitted.

how can you learn to use a tool that keeps changing unpredictably?

Well, I was assuming that what you learned about driving generation X of a tool would be at least mostly applicable to driving generation X+1 of it. That may be false, at least for some values of X.

The recent "opus 5 is worse than 4.8" suggests otherwise
Post reply on HN