Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Andrej Karpathy – It will take a decade to work through the issues with agents

dwarkesh.com

Post reply on HN
Print view View original

461–470 of 1001 posts

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#461

Post by timschmidt »

LarsDu88 wrote: post #443
ekjhgkejhgk wrote: post #134

Earlier quoted context omitted.

The interview which I've watched recently with Rich Sutton left me with the impression that AGI is not just a matter of adding more 9s. The interviewer had an idea that he took for granted: that to understand language you have to have a model of the world. LLMs seem to udnerstand language therefore they've trained a model of the world. Sutton rejected the premise immediately. He might be right in being skeptical here…

This world model talk is interesting, and Yann Lecunn has broached on the same topic, but the fact is there are video diffusion models that are quite good at representing the "video world" and even counterfactually and temporally coherently generating a representation of that "world" under different perturbations. In fact you can go to a SOTA LLM today, and it will do quite well at predicting the outcomes of basic co…

1000% this. I would only add this has been demonstrated explicitly with chess: https://adamkarvonen.github.io/machine_learning/2024/01/03/c...

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#462

Post by CamperBob2 »

Yizahi wrote: post #439
CamperBob2 wrote: post #382

Earlier quoted context omitted.

Let's get an English major to take a calculator to the International Math Olympiad, and see how that goes.

So a sign of AGI or intelligence on par with human is the ability to solve small generic math problems? And it still requires a handler human level intellinge to be paired with, to even start solving those math problems? Is that about right?

Not even close to right. First of all, the "small generic math problems" given at IMO are designed to challenge the strongest students in the world, and second, the recent results have been based on zero-shot prompts. The human operator did nothing but type in the questions and hit Enter.

If you do not understand the core concepts very well, by any rational definition of "understand," then you will not succeed at competitions like IMO. A calculator alone won't help you with math at this level, any more than a scalpel by itself would help you succeed at brain surgery.

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#463

Post by lm28469 »

simonsarris wrote: post #341
plaidfuji wrote: post #129

Earlier quoted context omitted.

Should probably just short nvidia

There is a wide space where LLMs and their offshoots make enormous productivity gains, while looking nothing like actual artificial intelligence (which has been rebranded AGI), and Nvidia turns out to have a justified valuation etc.

It's been three years now, where is it? Everyone on hn is now a 10x developers, where are all the new startups making $$$? Employees are 10x more productive, where are the 10x revenues? Or even 2x?

Why is growth over the last 3 years completely flat once you remove the proverbial AI pickaxes sellers?

What if all the slop generated by llms counterbalance any kind of productivity boost? 10x more bad code, 10x more spam emails, 10x more bots

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#464

Post by lm28469 »

mustaphah wrote: post #364

This aligns with METR's Time Horizons [1], the current SOTA "Moore's Law" for AI agents: - The length of tasks AI can complete doubles every ~7 months - In 2-4 years, AIs could autonomously complete week-long projects. - In under 10 years, they might handle month-long software or knowledge work. [1] https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...

It's like saying your newborn will have the same mass as earth in 50 years if he continues on his first month weight gain trajectory.

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#465

Post by themafia »

shdh wrote: post #433
themafia wrote: post #430

5 decades. You have one decade to clean up your power use problem. If you don't you will find yourself in the next AI winter.

Power use is less important than model capability AGI is either more scale or differing systems, or both They can always optimize for power consumption after AGI has been reached

> AGI is either more scale

So you plan to scale without increasing power usage. How's that?

> They can always optimize for power consumption after AGI has been reached

If you don't optimize power consumption you're going to increase surface area required to build it. There are hard physical limits having to do with signal propagation times.

You're ignoring the engineering entirely. The software is not hardly interesting or even evolving.

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#467

Post by ainch »

superconduct123 wrote: post #233

I always get a weird feeling when AI researchers and CS people start talking about comparisons between human brains and AI/computers Why is there a presumption that we (as people who have only studied CS) know enough about biology/neuroscience/evolution to make these comparisons/parallels/analogies? I enjoy the discussions but I always get the thought in the back of my head "...remember you're listening to 2 CS major…

There is a lot of overlap between AI and Neuroscience, especially among older researchers. For example Karpathy's PhD supervisor, Fei-Fei Li, researched vision in cat brains before working on computer vision, Demis Hassabis did his PhD in Computational Neuroscience, Geoff Hinton studied Psychology etc... There's even the Reinforcement Learning and Decision Making conference (RLDM - very cool!), which pairs Reinforcement Learning with neuro research and brings together people from both disciplines.

I suspect the average AI researcher knows much more about the brain than typical CS students, even if they may not have sufficient background to conduct research.

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#468

Post by bhelkey »

EA-3167 wrote: post #236
goalieca wrote: post #79

I remember attending a lecture from a famous quantum computing researcher in 2003. He said that quantum computing is 15-20 years away and then he followed up by saying that if he told anyone it was further away then he wouldn't get funding!

It's an excellent time-frame that sounds imminent enough to draw interest (and funding), but is distant enough that you can delay the promised arrival a few times in the span of a career before retiring. Fusion research lives and dies on this premise, ignoring the hard problems that require fundamental breakthroughs in areas such as materials science, in favor of touting arbitrary benchmarks that don't indicate real…

> companies will brag about limited roll-outs of niche cases in dry, flat, places that are easy to navigate

According to their website, Waymo offers autonomous rides to the general public in Austin, Atlanta, Phoenix, the San Francisco Bay Area, and Los Angeles [1].

* San Francisco is an extremely hilly city that gets a fair bit of fog.

* Los Angeles has notorious traffic and particularly aggressive drivers.

* Atlanta gets ~50 inches of rain a year, more than Seattle [2].

[1] https://waymo.com/faq/#:~:text=Where%20does%20Waymo%20operat...

[2] https://www.forbes.com/sites/marshallshepherd/2024/09/03/whi...

Re: Andrej Karpathy – It will take a decade to work through the issues with agents

#469

Post by rhetocj23 »

CamperBob2 wrote: post #462
Yizahi wrote: post #439

Earlier quoted context omitted.

So a sign of AGI or intelligence on par with human is the ability to solve small generic math problems? And it still requires a handler human level intellinge to be paired with, to even start solving those math problems? Is that about right?

Not even close to right. First of all, the "small generic math problems" given at IMO are designed to challenge the strongest students in the world, and second, the recent results have been based on zero-shot prompts. The human operator did nothing but type in the questions and hit Enter. If you do not understand the core concepts very well, by any rational definition of "understand," then you will not succeed at com…

It may be difficult for you to believe or digest, but this means nothing for actual innovation. Im yet to see the effects of LLMs send a shockwave in the real economy.

Ive actually hung around Olympiad level folks and unfortunately, their reach of intellect was limited in specific ways that didnt mean anything in regards to the real economy.

Post reply on HN