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

Will scaling work?

dwarkeshpatel.com

Post reply on HN
Print view View original

51–60 of 289 posts

Re: Will scaling work?

#51

Post by lkbm »

berniedurfee wrote: post #31

I think there’s a huge assumption here that more LLM will lead to AGI. Nothing I’ve seen or learned about LLMs leads me to believe that LLMs are in fact a pathway to AGI. LLMs trained on more data with more efficient algorithms will make for more interesting tools built with LLMs, but I don’t see this technology as a foundation for AGI. LLMs don’t “reason” in any sense of the word that I understand and I think the ab…

I guess it's an "assumption", but it's an assumption that's directly challenged in the article:

> But of course we don’t actually care directly about performance on next-token prediction. The models already have humans beat on this loss function. We want to find out whether these scaling curves on next-token prediction actually correspond to true progress towards generality.

And:

> Why is it impressive that a model trained on internet text full of random facts happens to have a lot of random facts memorized? And why does that in any way indicate intelligence or creativity?

And:

> So it’s not even worth asking yet whether scaling will continue to work - we don’t even seem to have evidence that scaling has worked so far.

Re: Will scaling work?

#52

Post by peteradio »

Der_Einzige wrote: post #33
nsagent wrote: post #22

Earlier quoted context omitted.

I mentioned this to another commenter as well: You might want to reconsider your stance on emergent abilities in LLMs considering the NeurIPS 2023 best paper winner is titled: "Are Emergent Abilities of Large Language Models a Mirage?" https://arxiv.org/abs/2304.15004 https://blog.neurips.cc/2023/12/11/announcing-the-neurips-20...

Papers which get accepted with honors are not necessarily more truthful than papers which have been rejected. Yann LeCunn goes on twitter like any other grad student around NeurIPS or ICML/ICMR and bitterly complains when one of his (many) papers is rejected. Whose more likely to be correct here? Yann LeCunn (the TOP nlp scholar in our field by citations, who does claim that most emergent capabilities are real in oth…

Yann LeCun through Meta is incentivized towards maximizing capital return based on local maxima. That is how all business works, there is not really a direct incentive to pushing boundaries beyond what can be immediately monetized.

Re: Will scaling work?

#53

Post by RandomLensman »

lumost wrote: post #43
enieslobby wrote: post #38

Earlier quoted context omitted.

Why do you think humans are basically evolved LLMs? Honest question, would love to read more about this viewpoint.

An LLM is simply a model which given a sequence, predicts the rest of the sequence. You can accurately describe any AGI or reasoning problem as an open domain sequence modeling problem. It is not an unreasonable hypothesis that brains evolved to solve a similar sequence modeling problem.

In the broader sense that is tricky as accurate prediction is not always the right metric (otherwise we'd still be using epicycles for the planets).

Re: Will scaling work?

#54

Post by jerpint »

slibhb wrote: post #49

The best analogy for LLMs (up to and including AGI) is the internet + google search. Imagine explaining the internet/google to someone in 1950. That person might say "Oh my god, everything will change! Instantaneous, cheap communication! The world's information available at light speed! Science will accelerate, productivity will explode!" And yet, 70 years later, things have certainly changed, but we're living in the…

The internet has allowed us to interact in ways that were inconceivable at the time; think communication and speed of information for one.

When agents start being more reliable I think we will start seeing applications we couldn’t possibly anticipate today

Re: Will scaling work?

#55

Post by nsagent »

Der_Einzige wrote: post #33
nsagent wrote: post #22

Earlier quoted context omitted.

I mentioned this to another commenter as well: You might want to reconsider your stance on emergent abilities in LLMs considering the NeurIPS 2023 best paper winner is titled: "Are Emergent Abilities of Large Language Models a Mirage?" https://arxiv.org/abs/2304.15004 https://blog.neurips.cc/2023/12/11/announcing-the-neurips-20...

Papers which get accepted with honors are not necessarily more truthful than papers which have been rejected. Yann LeCunn goes on twitter like any other grad student around NeurIPS or ICML/ICMR and bitterly complains when one of his (many) papers is rejected. Whose more likely to be correct here? Yann LeCunn (the TOP nlp scholar in our field by citations, who does claim that most emergent capabilities are real in oth…

Please at least read the paper before appealing to authority. It is a well designed set of experiments that clearly demonstrates that the notion of a "phase change" (rapid shift in capabilities) as a popularized by many people claiming emergence is actually a gradual improvement with more data.

But if you do want to appeal to Lecun as an authority, then maybe you'll accept that these (re)tweets that clearly indicate he finds the insights from the paper to be valid:

https://nitter.1d4.us/ylecun/status/1736479356917063847 https://nitter.1d4.us/rao2z/status/1736464000836309259 (retweeted)

As for Timnit, I think you have your timeline confused. Model cards are what put her on the map for most general NLP researchers, which predates her difficulties at Google.

2018: Model cards paper was put on arXiv https://arxiv.org/abs/1810.03993

2019: Major ML organizations start using model cards https://github.com/openai/gpt-2/blob/master/model_card.md

2020: Model cards become fairly standard https://blog.research.google/2020/07/introducing-model-card-...

Dec 2020: Timnit is let go from the ethics team at Google https://www.bbc.com/news/technology-55187611

EDIT:formatting

Re: Will scaling work?

#56

Post by spzb »

cortic wrote: post #42
enieslobby wrote: post #38

Earlier quoted context omitted.

Why do you think humans are basically evolved LLMs? Honest question, would love to read more about this viewpoint.

Look at a year old baby, there is no logic, no reasoning, no real consciousness, just basic algorithms and data input ports. It takes ten years of data sets before these emergent properties start to develop, and another ten years before anything of value can be output.

Have you ever met a baby? They're nothing like an LLM. For starters, they learn without using language. By one year old they've taught themselves to move around the physical world. They've started to learn cause and effect. They've learned where "they" end and "the rest of the world" begins. All an LLM has "learnt" is that some words are more likely to follow others.

Re: Will scaling work?

#57

Post by beepbooptheory »

mgaunard wrote: post #37

I think the more interesting question is how long will people cling to the illusion that LLMs will lead us to AGI? Maintaining the illusion is important to keep the money flowing in.

While this is certainly true, I think we can't ignore the intense enthusiasm and faith of a large cohort of our peers (or, you know, HN commenters) who believe this to be The Way, and are not necessarily stakeholders in any meaningful sense. Just look at some of the responses even in this thread. It feels like some people just need this, and respond to balanced skepticism as Alyosha does to his brother Ivan.

In part, whether conscious or not, people see the bright future of LLMs as a kind of redemption for the world so far wrought from a Silicon Valley ideology; its almost too on-the-nose the way chatgpt "fixes" internet search.

But on a deeper level, consider how many hn posts we saw before chatgpt that were some variation of "I have reached a pinnacle of career accomplishment in the tech world, but I can't find meaning or value in my life." We don't seem to see those posts quite as much with all this AI stuff in the air. People seem to find some kind of existential value in the LLMs, one with an urgency that does not permit skepticism or critique.

And, of course, in this thread alone, there is the constant refrain: "well, perhaps we are large language models ourselves after all..." This reflex to crude Skinnerism says a lot too: there are some that, I think, seek to be able to conquer even themselves; to reduce their inner life to python code and data, because it is something they can know and understand and thus have some kind of (sense) of control or insight about it.

I don't want to be harsh saying this, people need something to believe in. I just think we can't discount how personal all this appears to be for a lot of regular, non-AI-CEO people. It is just extremely interesting, this culture and ideology being built around this. To me it rivals the LLMs themselves as a kind fascinating subject of inquiry.

Re: Will scaling work?

#58

Post by visarga »

berniedurfee wrote: post #50
cortic wrote: post #42

Earlier quoted context omitted.

Look at a year old baby, there is no logic, no reasoning, no real consciousness, just basic algorithms and data input ports. It takes ten years of data sets before these emergent properties start to develop, and another ten years before anything of value can be output.

I strongly disagree. Kids, even infants, show a remarkable degree of sophistication in relation to an LLM. I admit that humans don’t progress much behaviorally, outside of intellect, past our teen years; we’re very instinct driven. But still, I think even very young children have a spark that’s something far beyond rote token generation. I think it’s typical human hubris (and clever marketing) to believe that we can…

Humans are not very smart, individually, and over a single lifetime. We become smart as a species in tens of millennia of gathering experience and sharing it through language.

What LLMs learn is exactly the diff between primitive humans and us. It's such a huge jump a human alone can't make it. If we were smarter we should have figured out the germ theory of disease sooner, as we were dying from infections.

So don't praise the learning abilities of little children, without language and social support they would not develop very much. We develop not just by our DNA and direct experiences but also by assimilating past experiences through language. It's a huge cache of crystallized intelligence from the past, without which we would not rule this planet.

That's also why I agree LLMs are stalling because we can't quickly scale a few more orders of magnitude the organic text inputs. So there must the a different way to learn, and that is by putting AI in contact with environments and letting it do its own actions and learn from its mistakes just like us.

I believe humans are "just" contextual language and action models. We apply language to understand, reason and direct our actions. We are GPTs with better feedback from outside, and optimized for surviving in this environment. That explains why we need so few samples to learn, the hard work has been done by many previous generations, brains are fit for their own culture.

So the path forward will imply creating synthetic data, and then somehow evaluating the good from the bad. This will be task specific. For coding, we can execute tests. For math, we can use theorem provers to validate. But for chemistry we need simulations or labs. For physics, we need the particle accelerator to get feedback. But for games - we can just use the score - that's super easy, and already led to super-human level players like AlphaZero.

Each topic has its own slowness and cost. It will be a slow grind ahead. And it can't be any other way, AI and AGI are not magic. They must use the scientific method to make progress just like us.

Re: Will scaling work?

#59

Post by arketyp »

slibhb wrote: post #49

The best analogy for LLMs (up to and including AGI) is the internet + google search. Imagine explaining the internet/google to someone in 1950. That person might say "Oh my god, everything will change! Instantaneous, cheap communication! The world's information available at light speed! Science will accelerate, productivity will explode!" And yet, 70 years later, things have certainly changed, but we're living in the…

My take on this is that much of work and problem solving is about understanding the problem. So I think human abilities will remain the bottleneck. I pose this thought experiment: Is it possible to design an AI system for a monkey which gives it super-monkey abilities?

Re: Will scaling work?

#60

Post by bee_rider »

slibhb wrote: post #49

The best analogy for LLMs (up to and including AGI) is the internet + google search. Imagine explaining the internet/google to someone in 1950. That person might say "Oh my god, everything will change! Instantaneous, cheap communication! The world's information available at light speed! Science will accelerate, productivity will explode!" And yet, 70 years later, things have certainly changed, but we're living in the…

The internet did change things pretty dramatically.

Productivity at information communication tasks just isn’t the entire economy.

I think we are massively more productive. Some of the biggest new companies are ad companies (Google, Facebook), or spend a ton of their time designing devices that can’t be modified by their users (Apple, Microsoft). Even old fashioned companies like tractor and train companies have time to waste on preventing users from performing maintenance. And then the economy has leftover effort to jailbreak all this stuff.

We’re very productive, we’ve just found room for unlimited zero or negative sum behavior.

Post reply on HN