[antiquark]

I doubt that word models can lead to world models. To quote Yann LeCun:

"The vast majority of our knowledge, skills, and thoughts are not verbalizable. That's one reason machines will never acquire common sense solely by reading text."

https://twitter.com/ylecun/status/1368235803147649028

[moffkalast]

That just seems like an unfounded hot take. Of course we can explain most of our knowledge, skills, and thoughts in words, that's how we don't lose everything when the next generation comes around lol. It's the core reason we're different from animals.

Now sure you can't describe qualia, but that's basically a subjective artefact of how we sense the world and (to add another unfounded hot take) likely not critical to have an understanding of it on a physical level.

[dimatura]

I disagree that this is an "unfounded hot take". It's far from a rare opinion on cognitive science, and if I had to guess it's probably the mainstream opinion (I can't really back that up with citations because I haven't followed the field closely in the last decade). And for what it's worth, I agree with Yann, although I have to admit that LLMs work far better than I would've guessed.

It's a topic that's too large for an HM comment, but "explaining" things in words comes after the fact, and mostly limited to a small subset of our experience and skillset that is amenable to it.

Note that humans are animals too, btw. And conversely, I would consider nonverbal people as humans as well.

[moffkalast]

Well I admit I used to be of a similar opinion as well, but seeing this explosion unravel over the past few months has me convinced that it's can't possibly be right, at least not to any degree that objectively matters.

Perhaps language is the wrong term to use, since it's not what LLMs are really about. They're about text. There are very few things that cannot be expressed as text, albeit in unconventional ways like base64. Being opaque to humans doesn't mean that with enough data a neural net can't be taught to "see" images that way or "hear" sound files for example. If the original assertion is true, then there must be some kind of universal barrier to skills that cannot be expressed in text. That sounds completely crazy to me, since we humans are also likely just organic data that could be expressed as text with some encoding. The main problem is interfacing with it in some way that's actually useful, which is the extremely hard part.

Another thing to consider is that with a formalized enough language (i.e. a programming language) one can be far more exact in explaining things accurately than any natural language with its cultural specifics and inferred nonsense. That's probably why LLMs designed as coding models first and foremost usually outperform those that aren't in solving unrelated arbitrary problems.

> Note that humans are animals too, btw. And conversely, I would consider nonverbal people as humans as well.

Humans are animals in the biological sense, yes. But very much not in the societal and skill-transferring sense.

[dimatura]

Well, that idea is one of the motivations behind the paper, which is itself a throwback to earlier ideas about the "language of thought", a hypothetical language (certainly not ordinary natural language, and probably more like a programming language). But adding a few twists such as the probabilistic part and of course the whole machinery of LLMs, and more emphasis on sensory grounding. I think it's a very interesting approach from a researcher I respect, but obviously don't know if it'll pan out.

[pphysch]

> Of course we can explain most of our knowledge, skills, and thoughts in words, that's how we don't lose everything when the next generation comes around lol.

I would wager if you put a newborn human to be raised in the absence of any physical human contact, but somehow taught them to read/write, and gave them access to a universal corpus (text only, no audio/video), or heck, even internet access with `curl`, and lastly dropped them into the "real world" at age 25, they would be utterly incapable of performing, say, a basic service job at a restaurant.

Words help us symbolize and reason about our sense experiences, but they are not a substitute for them.

[esafak]

Sure. Reading about colors will tell you nothing about them until you can see a depiction of them attached to their names. Same with all the other senses.

[NoMoreNicksLeft]

> Of course we can explain most of our knowledge, skills, and thoughts in words,

This is either some profound miscomprehension of just how many of your skills and thoughts are inexpressible in words, or some statement of how profoundly shallow your skills and thoughts actually are.

[FrustratedMonky]

>"solely by reading text".

Of course, that does leave the door Open, that when these models are put in a physical real body, a robot, and have to interact with the world, then maybe they can gain that "common sense".

This doesn't mean a silicon based AI can't become conscious of skills that are hard to verbalize. Just that they don't yet have all the same inputs that we have. And when they do, and they have internal thoughts, they will have the same difficulty verbalizing them that we do.

[stevenhuang]

Yann LeCun has a vested interest in downplaying LLM emergent abilities.

His research at meta is in the analytic approach to machine learning. As result he is very unabashed in expressing distaste of ML approaches that don't align with his research.

Really, there is no larger sore loser than LeCun in internalizing the bitter lesson. Quoting him without this context is being deliberately misleading.

[valine]

What concepts exactly can’t be verbalized? All of our serialized file formats fall under the umbrella of “words”. GPT4 can draw images by outputting SVGs for example.

[antiquark]

> What concepts exactly can’t be verbalized?

I would like to explain, but I can't quite put it into words...

:)

[valine]

Trying drawing a picture of it and save it as an SVG.