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

Please ignore the deluge of complete nonsense about Q*

twitter.com

Post reply on HN
Print view View original

71–80 of 208 posts

Re: Please ignore the deluge of complete nonsense about Q*

#71

Post by naasking »

qualifiedai wrote: post #46
hackinthebochs wrote: post #38

Earlier quoted context omitted.

His arguments in defense of barreling forward with AI are terrible. They have zero chance to convince someone who doesn't share his intuitions/interests. For example: https://twitter.com/ylecun/status/1718764953534939162 How easily smart people convince themselves of what they want to be true with zero self-awareness makes me much more fearful of what's to come. >other side gives, imho, convoluted arguments and asks…

The default in science is that the side arguing a point has the burden of proving it correct. Not asking the other side to prove them wrong.

> The default in science is that the side arguing a point has the burden of proving it correct

Right, and Yann is arguing the point that AI and LLMs are not or will not be dangerous. Where's his proof? As the parent posters have said, he has none.

Re: Please ignore the deluge of complete nonsense about Q*

#72

Post by the_gipsy »

bryan0 wrote: post #68

> One of the main challenges to improve LLM reliability is to replace Auto-Regressive token prediction with planning. Anyone have a good source for learning more about this topic?

It's just making a plan with auto-regressive token prediction, that's all.

Re: Please ignore the deluge of complete nonsense about Q*

#74

Post by loveparade »

cgearhart wrote: post #70

Does anyone have any references to published works discussing the idea he’s pushing here that all the top labs are trying to replace next token prediction with planning? (Note: I’m not interested in using a token-prediction LLM to solve planning problems; I’m interested in work that tries to replace token sampling with action sampling by formulating the text completion problem as a planning domain.) Or is that not wh…

If you do search for Reinforcement Learning based decoding in language models (or neural machine translation) you'll find hundreds of papers. It's a super common research topic that has been around forever, way before LLMs were a "thing". Just follow the reference from there. Nothing new here, really.

But doing it on the scale of something like GPT-4 is only something OAI can do, and I don't doubt that there's some breakthrough there. But IMO it's likely that the breakthrough isn't algorithmic, but a result of scale, like most of the other OAI breakthroughs.

Re: Please ignore the deluge of complete nonsense about Q*

#75

Post by kristopolous »

There's so much attention on this field right now (something like ~100 papers hitting arxiv a day). Is the possibility of being blind-sided by a "next generation" technique really plausible?

Isn't it more plausible that someone is just piecing together a couple key public papers in a novel way?

Re: Please ignore the deluge of complete nonsense about Q*

#76

Post by qualifiedai »

naasking wrote: post #71
qualifiedai wrote: post #46

Earlier quoted context omitted.

The default in science is that the side arguing a point has the burden of proving it correct. Not asking the other side to prove them wrong.

> The default in science is that the side arguing a point has the burden of proving it correct Right, and Yann is arguing the point that AI and LLMs are not or will not be dangerous. Where's his proof? As the parent posters have said, he has none.

Show me LLM that has reached a level of automony and self sufficiency of a cat (Yann's example) or even of a simple beattle.

LLMs are tools in amplifying individual human intelligence, 100% of automony and will come from their user (human).

Also, even as tools they have fundamental limitations which stem from their autoregressive nature

Re: Please ignore the deluge of complete nonsense about Q*

#77

Post by CSMastermind »

bbor wrote: post #58
qualifiedai wrote: post #46

Earlier quoted context omitted.

The default in science is that the side arguing a point has the burden of proving it correct. Not asking the other side to prove them wrong.

A nice thought but not very helpful. What’s the null hypothesis here?

The null hypothesis is that technological advancement will be a net positive for humanity, as all previous technological advancements have been.

Re: Please ignore the deluge of complete nonsense about Q*

#78

Post by melagonster »

ren_engineer wrote: post #33

The leak about Q* feels like an olive branch to let the former OpenAI board and Ilya save a bit of face, probably part of the terms for Sam coming back plus it distracts from all the drama and puts a positive spin on things. >"The board didn't handle things well, but they were right to be concerned because OpenAI did have some sort of research breakthrough" Not coincidence that this leaks after Sam comes back, rather…

hope openAI prepared some physical mechanisms to make sure safety of human.

p.s., maybe bomb us a good idea.

Re: Please ignore the deluge of complete nonsense about Q*

#79

Post by lucubratory »

concordDance wrote: post #50
qualifiedai wrote: post #13

Earlier quoted context omitted.

On the contrary, he gives good arguments about why open is safer and closed is more dangerous whereas other side gives, imho, convoluted arguments and asks for them to be proven wrong (as opposed to trying to prove themselves right).

I've read a fair bit of LeCunn and frankly have been very unimpressed. He is, as far as I can tell, the only big AI name who doesn't think that if we somehow managed to make an actual superintelligent AI it wouldn't be dangerous.

And his reasoning for it is absurd, a combination of evolutionary psychology (not the most scientifically rigorous field to begin with and not even his field) and the belief that no one would choose to make an unsafe AI system, while simultaneously arguing for open sourcing everything so that anyone on the planet can do so if they want.

Re: Please ignore the deluge of complete nonsense about Q*

#80

Post by born-jre »

Looks like he is not saying it cz he has inside knowledge, instead he is just saying his usual I-JEPA line. About which, the best way to convince people is to give them something they can evaluate/get their hands dirty. I-JEPA for now still looks hand wavy / incomplete.
Post reply on HN