AI/LLMs Can LLMs Think? - The Debate

[Umbran]

I also wonder how much this is going to throw cold water on the usage of things like Codex if you know that the company will throw its unlimited compute at it once you get close enough so that they can steal the credit and a headline.

Imagine the blinders you have to be wearing to not see it coming.

Like, dudes, the machines you are using are built upon a foundation of taking whatever data they wanted without asking, giving credit, or paying for rights!

 

[The Firebird]

I think it's important to take a step back to appreciate exactly what the Millennium problems represent and all the human capital that has been spent working on them over the years. While the prizes were formally established in 2000, many of the problems originated much earlier. For many, if not all of these problems, entire careers have been spent on partial progress, new techniques, and failed approaches, without meaningfully getting closer to a solution.

The Clay Mathematics Institution - Millennium Problems

Likewise, the point Tristan made prior to the confusion with OpenAI:

"I had planned to say on announcing our work that the results are not the important thing. Rather the important thing is instead the significance that a mathematician and an LLM model can now do all this work in a month. The significance of this with respect to the way we train students, assign credit, referee, and decide what is worth one human life’s attention cannot be understated. This is a a Deep Blue-Kasparov moment. The community needs to have serious and unhurried discussion about where to go from here"

Trust the experts, you know

 

[The Firebird]

I stumbled on this video which I think would be of interest here.

The TL;DR: the speaker is a theoretical physicist who transitioned to AI and now works for Google/Deepmind. His point is that in six years AI/LLM went from a preschooler level to consistently solving non-trivial science problems at the level of good graduate students on their own. Now we are moving into a regime where, under human supervision, they are performing novel math research. At the same time, the cost/resources required for a given level of intelligence have decreased by several orders of magnitude, and he doesn't believe both trends are not going to stop anytime soon.

Right, the cost point is a big one. Noam Brown made a good argument: when o3 was announced (under 2 years ago!) it took $500,000 to score 87.5% on the ARC-AGI-1 benchmark. Astra has superior performance for $20. If the costs keep dropping so precipitously...

 

[abazaba]

I know what comes to mind....

A song, by the way, which is in the public domain...

'An error occurred. Please try again later' ? Such a strange name for a song.

 

[Umbran]

'An error occurred. Please try again later' ? Such a strange name for a song.

View attachment 447573

Weird it should be fixed now.

 

[Umbran]

Likewise, the point Tristan made prior to the confusion with OpenAI:

"...The community needs to have serious and unhurried discussion about where to go from here"

The community must collectively be willing to say, "No," to have an unhurried discussion. If they cannot do that, the discussion will be rushed.

 

[Stalker0]

Yes, but it wasn't free. Reportedly, it took them $22 million dollars of computing time to try to scoop the human mathematicians.

$22 million.

At a salary of $125k, that would be... 176 mathematicians for a year. Or 17 mathematicians for a decade.

Was that one proof worth more to human knowledge than the output of 17 mathematicians for a decade?

1) it might be actually.

2) ignoring 1, your trying to make a straight financial comparison but are ignoring the benefit of not just solving thr problem, but in the system and teams learning how to solve such problems better.

When technology is advanced it’s never “financially profitable” at the onset, but it improves with iteration becoming faster, cheaper, better until it hits that threshold.

The tech is now at the point where it’s good enough to do the job…but if it’s still too expensive than the next step is to make it more efficient. That’s the same cycle we apply to any technology

 

[Stalker0]

Unfortunately on to the main topic of this thread, the Navier-Stokes situation is just not a good topic for the overall conversation due to the scandel.

Right now its too hard to parse out how much the AI did on its own versus just walking the path the mathematician's work had already laid out...so I feel any discussion on the AI's thinking will just get quagmired into that debate rather than the elements of thinking itself.

 

[The Firebird]

Unfortunately on to the main topic of this thread, the Navier-Stokes situation is just not a good topic for the overall conversation due to the scandel.

Right now its too hard to parse out how much the AI did on its own versus just walking the path the mathematician's work had already laid out...so I feel any discussion on the AI's thinking will just get quagmired into that debate rather than the elements of thinking itself.

Even in the case it was taken from stolen codex chats, those are llms and it was LLM assisted. To such an extent that the author, who it would have been stolen from, calls it a Deep Blue moment.

That doesn't mean the LLM is thinking anymore than stockfish is...but it does mean a lot of things we consider to require thought are actually quite amendable to calculation.

 

[Coroc]

Imagine the blinders you have to be wearing to not see it coming.

Like, dudes, the machines you are using are built upon a foundation of taking whatever data they wanted without asking, giving credit, or paying for rights!

Thereby suggesting that usage of most features is free ;P