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

231–240 of 343 posts

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

#231

Post by tgma »

chongli wrote: post #221
csomar wrote: post #213

Earlier quoted context omitted.

All of these "experiences" are encoded in your brain as electricity. So "text" can encode them, though English words might not be the proper way to do it.

No, text can only refer to them. 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.

> 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 be reduced to electrical signals after all). The only question is how much storage you need for that encoding to get what precision. However, the latter statement, if true, is simply constrained by your input device to the brain, i.e. you cannot transfer your encoding to the hardware in this case a human brain via reading or listening. There could be higher bandwidth interfaces like neuralink that may do that to human brain and in the case of AI, an auxiliary device might not be needed and the encoding would be directly mmap'd.

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

#232

Post by andsoitis »

codenlearn wrote: post #138
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…

Doesn't Language itself encode multimodal experiences? Let's take this case write when we write text, we have the skill and opportunity to encode the visual, tactile, and other sensory experiences into words. and the fact is llm's trained on massive text corpora are indirectly learning from human multimodal experiences translated into language. This might be less direct than firsthand sensory experience, but potentia…

Some aspects of experience— e.g. raw emotions, sensory perceptions, or deeply personal, ineffable states—often resist full articulation.

The taste of a specific dish, the exact feeling of nostalgia, or the full depth of a traumatic or ecstatic moment can be approximated in words but never fully captured. Language is symbolic and structured, while experience is often fluid, embodied, and multi-sensory. Even the most precise or poetic descriptions rely on shared context and personal interpretation, meaning that some aspects of experience inevitably remain untranslatable.

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

#233

Post by tsimionescu »

csomar wrote: post #213
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…

All of these "experiences" are encoded in your brain as electricity. So "text" can encode them, though English words might not be the proper way to do it.

We don't know how memories are encoded in the brain, but "electricity" is definitely not a good enough abstraction.

And human language is a mechanism for referring to human experiences (both internally and between people). If you don't have the experiences, you're fundamentally limited in how useful human language can be to you.

I don't mean this in some "consciousness is beyond physics, qualia can't be explained" bullshit way. I just mean it in a very mechanistic way: language is like an API to our brains. The API allows us to work with objects in our brain, but it doesn't contain those objects itself. Just like you can't reproduce, say, the Linux kernel just by looking at the syscall API, you can't replace what our brains do by just replicating the language API.

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

#234

Post by chongli »

tgma wrote: post #231
chongli wrote: post #221

Earlier quoted context omitted.

No, text can only refer to them. 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.

> 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.

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

#235

Post by flamedoge »

Lerc wrote: post #50
seanhunter wrote: post #9

Earlier quoted context omitted.

> The problem with LLMs is that the output is inherently stochastic - i.e there isn't a "I don't have enough information" option. This is due to the fact that LLMs are basically just giant look up maps with interpolation. I don't think this explanation is correct. The input to the decoder at the end of all the attention heads etc (as I understand it) is a probability distribution over tokens. So the model as a whole…

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 literally doesn't know how to handle 'I don't know' and needs to be taught. fascinating.

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

#236

Post by satellite2 »

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…

So blind people aren't generally sentient?

(I'm obviously exaggerating a bit for the sake of the argument, but the point stands. Multimodality should not be a prerequisite to AGI)

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

#237

Post by Terr_ »

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.

> text definitely encodes multimodal experiences

Perhaps, but only in the same sense that brown and green wax on paper "encodes" an oak tree.

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

#238

Post by somenameforme »

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…

Is that what he's arguing? My perspective on what he's arguing is that LLMs effectively rely on a probabilistic approach to the next token based on the previous. When they're wrong, which the technology all but ensures will happen with some significant degree of frequency, you get cascading errors. It's like in science where we all build upon the shoulders of giants, but if it turns out that one of those shoulders was simply wrong, somehow, then everything built on top of it would be increasingly absurd. E.g. - how the assumption of a geocentric universe inevitably leads to epicycles which leads to ever more elaborate, and plainly wrong, 'outputs.'

Without any 'understanding' or knowledge of what they're saying, they will remain irreconcilably dysfunctional. Hence the typical pattern with LLMs:

---

How do I do [x]?

You do [a].

No that's wrong because reasons.

Oh I'm sorry. You're completely right. Thanks for correcting me. I'll keep that in mind. You do [b].

No that's also wrong because reasons.

Oh I'm sorry. You're completely right. Thanks for correcting me. I'll keep that in mind. You do [a].

FML

---

More advanced systems might add a c or a d, but it's just more noise before repeating the same pattern. Deep Seek's more visible (and lengthy) reasoning demonstrates this perhaps the most clearly. It just can't stop coming back to the same wrong (but statistically probable) answer and so ping-ponging off that (which it at least acknowledges is wrong due to user input) makes up basically the entirety of its reasoning phase.

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

#239

Post by thousand_nights »

satellite2 wrote: post #236
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…

So blind people aren't generally sentient? (I'm obviously exaggerating a bit for the sake of the argument, but the point stands. Multimodality should not be a prerequisite to AGI)

blind people still have other senses including touch which gives them a size reference they can compare to. you can feel physical objects to gain an understanding of their size.

the LLM is more like a brain in a vat with only one sensory input - a stream of text

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

#240

Post by Lerc »

flamedoge wrote: post #235
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 literally doesn't know how to handle 'I don't know' and needs to be taught. fascinating.

I think it would be more accurate to say that after fine tuning on a series of questions with answers that it thinks that you don't want to hear "I don't know"
Post reply on HN