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

Ask HN: Any insider takes on Yann LeCun's push against current architectures?

news.ycombinator.com

Post reply on HN
Print view View original

271–280 of 343 posts

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#271

Post by anonymoushn »

vessenes wrote: post #27
probably_wrong wrote: post #23

I haven't read Yann Lecun's take. Based on your description alone my first impression would be: there's a paper [1] arguing that "beam search enforces uniform information density in text, a property motivated by cognitive science". UID claims, in short, that a speaker only delivers as much content as they think the listener can take (no more, no less) and the paper claims that beam search enforced this property at ge…

I believe he’s talking about some sort of ‘energy as measured by distance from the models understanding of the world’ as in quite literally a world model. But again I’m ignorant, hence the post!

This sort of measure is a decent match for BPB though. BPB=-log(document_probability)/document_length_bytes and perplexity=e^(BPB*document_length_bytes/document_length_tokens). We already train models by minimizing perplexity, and model outputs are already those that are high probability. Though like with EBMs, figuring out outputs with even higher probability would require an expensive search step.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#272

Post by yahoozoo »

kadushka wrote: post #262
yahoozoo wrote: post #256

Earlier quoted context omitted.

Does it really matter? At the end of the day, all the modalities and their architectures boil down to matrices of numbers and statistical probability. There’s no agency, no soul.

At the end of the day, all modalities boil down to patterns of electrical activity in your brain.

The brain is the important part. The electricity just keeps it going. And it’s more than numerical matrices.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#273

Post by mietek »

chongli wrote: post #234
tgma wrote: post #231

Earlier quoted context omitted.

> There is not a text on this planet that encodes what the heat of the sun feels like on your skin. > A person who had never been outdoors could never experience that sensation by reading text. I don't think the latter implies the former as obviously as you make it to be. Unless you believe in some sort of metaphysical description of human, you can certainly encode the feeling (as mentioned in another comment it will…

Electrical signals are not the same as subjective experiences. While a machine may be able to record and play back these signals for humans to experience, that does not imply that the experiences themselves are recorded nor that the machine has any access to them. A deaf person can use a tape recorder to record and play back a symphony but that does not encode the experience in any way the deaf person could share.

That’s some strong claims, given that philosophers (e.g. Chalmers vs Dennett) can’t even agree whether subjective experiences exist or not.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#274

Post by fluidcruft »

spyder wrote: post #181
ThinkBeat wrote: post #146

Earlier quoted context omitted.

No. > The sun feels hot on your skin. No matter how many times you read that, you cannot understand what the experience is like. > You can read a book about Yoga and read about the Tittibhasana pose But by just reading you will not understand what it feels like. And unless you are in great shape and with greate balance you will fail for a while before you get it right. (which is only human). I have read what shooting…

> No. Huh, text definitely encodes multimodal experiences, it's just not as accurate and as rich encoding as the encodings of real sensations.

Language encodes what people need it to encode to be useful. I heard of an example of colors--there are some languages that don't even have a word for blue.

https://blog.duolingo.com/color-words-around-the-world/

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#275

Post by Lerc »

cruffle_duffle wrote: post #265
Lerc wrote: post #50

Earlier quoted context omitted.

I feel like we're stacking naive misinterpretations of how LLMs function on top of one another here. Grasping gradient descent and autoregressive generation can give you a false sense of confidence. It is like knowing how transistors make up logic gates and believing you know more than CPU design than you actually do. Rather than inferring from how you imagine the architecture working, you can look at examples and co…

> It would be quite easy to detect (and exploit) behaviour that decided to use a vowel word just because it somewhat arbitrarily used an 'an'. That is a very interesting observation! Doesn’t that internal state get blown away and recreated for every “next token”? Isn’t the output always the previous context plus the new token, which gets fed back and out pops the new token? There is no transfer of internal state to t…

>Doesn’t that internal state get blown away and recreated for every “next token”

That is correct. When a model has a good idea of the next 5 words, after it has emitted the first of those 5 most architectures make no further use of the other 4 and regenerate likely the same information again in the next inference cycle.

There are architectures that don't discard all that information but the standard LLM has generally outperformed them, for now.

There are interesting philosophical implications if LLMs were to advance to a level to be considered sentient. Would it not be constantly creating and killing a thinking being for every token. On the other hand if context is considered memory, perhaps continuity of identity is based upon memory and all that other information are simply forgotten idle thoughts. We have no concept of what our previous thoughts were except from our memory. Is that not the same.

Sometimes I wonder if some of the resistance to AI is because it can do things that we think requires abilities that we would like to believe that we possess ourselves, and showing that they are not necessary creates the possibility that we might not have have those abilities.

There was a great observation recently in an interview (I forget the source, but the interviewer's last name was Bi) that some of the discoveries that met the most resistance in history such as the Earth orbiting the Sun, or Darwin's theory of evolution were similar in that they implied that we are not a unique special case.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#276

Post by rkp8000 »

pessimizer wrote: post #207
ninetyninenine wrote: post #204

Earlier quoted context omitted.

LLM is just the name. You can encode anything into the "language" including pictures video and sound.

I've always been wondering if anyone is working on using nerve impulses. My first thought when transformers came around was if they could be used for prosthetics, but I've been too lazy to do the research to find anybody working on anything like that, or to experiment myself with it.

There are a few folks working on this in neuroscience, e.g. training transformers to "decode" neural activity (https://arxiv.org/abs/2310.16046). It's still pretty new and a bit unclear what the most promising path forward is, but will be interesting to see where things go. One challenge that gets brought up a lot is that neuroscience data is often high-dimensional and with limited samples (since it's traditionally been quite expensive to record neurons for extended periods), which is a fairly different regime from the very large data sets typically used to train LLMs, etc.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#277

Post by amw-zero »

jcims wrote: post #203
bravura wrote: post #132

Okay I think I qualify. I'll bite. LeCun's argument is this: 1) You can't learn an accurate world model just from text. 2) Multimodal learning (vision, language, etc) and interaction with the environment is crucial for true learning. He and people like Hinton and Bengio have been saying for a while that there are tasks that mice can understand that an AI can't. And that even have mouse-level intelligence will be a br…

Over the last few years I’ve become exceedingly aware at how insufficient language really is. It feels like a 2D plane and no matter how many projections you attempt to create from it, they are ultimately limited in the fidelity of the information transfer. Just a lay opinion here but to me each mode of input creates a new, largely orthogonal dimension for the network to grow into. The experience of your heel slippin…

This exactly describes my intuition as well. Language is limited by its representation, and we have to jam so many bits of information into one dimension of text. It works well enough to have a functioning society, but it’s not very precise.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#278

Post by jononor »

chriskanan wrote: post #248
jononor wrote: post #137

Earlier quoted context omitted.

There are many roadblocks to continual learning still. Most current models and training paradigms are very vulnerable to catastrophic forgetting. And are very sample inefficient. And we/the methods are not so good at separating what is "interesting" (should be learned) vs "not". But this is being researched, for example under the topic of open ended learning, active inference, etc.

As a leader in the field of continual learning, I somewhat agree, but I'd say that catastrophic forgetting is largely resolved. The problem is that the continual learning community largely has become insular and is mostly focusing on toy problems that don't matter, where they will even avoid good solutions for nonsensical reasons. For example, reactivation / replay / rehearsal works well for mitigating catastrophic f…

Thanks a lot for this paper and the ones you shared deeper in the thread!

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#279

Post by kryogen1c »

ninetyninenine wrote: post #204
jcims wrote: post #203

Earlier quoted context omitted.

Over the last few years I’ve become exceedingly aware at how insufficient language really is. It feels like a 2D plane and no matter how many projections you attempt to create from it, they are ultimately limited in the fidelity of the information transfer. Just a lay opinion here but to me each mode of input creates a new, largely orthogonal dimension for the network to grow into. The experience of your heel slippin…

LLM is just the name. You can encode anything into the "language" including pictures video and sound.

> You can encode anything into the "language

Im just a layman here, but i don't think this is true. Language is an abstraction, an interpreative mechanism of reality. A reproduction of reality, like a picture, by definition holds more information than it's abstraction does.

Re: Ask HN: Any insider takes on Yann LeCun's push against current architectures?

#280

Post by emrah »

chriskanan wrote: post #246

A lot of the responses seem to be answering a different question: "Why does LeCun think LLMs won't lead to AGI?" I could answer that, but the question you are asking is "Why does LeCun think hallucinations are inherent in LLMs?" To answer your question, think about how we train LLMs: We have them learn the statistical distribution of all written human language, such that given a chunk of text (a prompt, etc.) it then…

Would you care to post your ideas somewhere online so others can read, critique, try etc?
Post reply on HN