AI/LLMs Can LLMs Think? - The Debate

[Umbran]

1) it might be actually.

Bold statement. Care to back it up?

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.

Yes, well, then let's also include the number of new mathematicians and teh amount of mathematical literacy they will create in their ten years, shall we?

The tech is now at the point where it’s good enough to do the job…

Except... it isn't. That's been covered - the technolgody didn't come up with how to solve the problem - it only executed on a known path...

... in an apparent attempt to claim they are good enough, and to steal glory.

 

[Umbran]

Thereby suggesting that usage of most features is free ;P

Except it isn't. That is an illusion.

The cost of those features are dumped out on the public (in the form of loss of income for creators, and financial risk to the general public), rather than absorbed by the company.

This is like saying burning fossil fuels is "cheap", by neglecting the billions and billions of dollars of damage incurred to the rest of us via environmental impacts of letting carbon loose in the atmosphere.

 

[Stalker0]

Bold statement. Care to back it up?

No more than you would like to back up yous I imagine. Calculating the financial benefit of knowledge acquisition has always been terribly complicated because its hard for us to predict what knowledge will actually be useful. Lasers went from pet project fringe sceience to one of the most important underpinnings of modern technology. Meanwhile, we have spent decades on string theory without a lot of commercial application.

So would 176 man-years of mathematican time bore better fruit than the solution to this Navier-Stokes millenium problem? Maybe, maybe not.

Yes, well, then let's also include the number of new mathematicians and teh amount of mathematical literacy they will create in their ten years, shall we?

Of course.

Except... it isn't. That's been covered - the technolgody didn't come up with how to solve the problem - it only executed on a known path...

... in an apparent attempt to claim they are good enough, and to steal glory.

Yes to my recent point...unfortunately this particular example is not the best when it comes to saying "how much is the AI thinking" because its still hard to parse out what the AI did on its own versus just walked the path already laid before it. Its very unfortunate, so we have to look to other examples for the debate.




Overall, I have seen you bring up this point a few times regarding the "cost" of AI intellligence in terms of money, resources, etc. Your point being that AI is simply too expensive which therefore nullifies it as a viable option for thought expansion. My point is, all new technologies start that way. We are already seeing signs that AI can solve problems that traditional have taken very intelligent very educated people to solve. And the difference is, the AI is getting smarter every year, whereas mathematicians still require decades of schooling to get to the point where they contribute meaningfully to their areas. If AI is capable of doing that work, than it becomes an efficiency problem....we need more intelligence per dollar. That's an engineering and computer science problem, and we have traditionally been very good at taking new technologies and making them more efficient once they hit the threshold of true utility. there is nothing to me that AI is any different in this repect than the hundreds of ground breaking techs that have come before it.

Now I can respect that we have taken such an "inefficienT" technology and expanded it way too quickly before its proven itself, that I could fully respect and agree with. But it doesn't detract from the more fundamental point that the AI IS able to do work that even a few years ago would have been a pipe dream...work that has traditionally been considered very "thoughful" work.

 

[The Firebird]

OpenAI: "In addition, since the completion of Navier-Stokes, we have made substantial progress on another Millennium Prize problem. We are working through how to share these results thoughtfully.”

Link

In the same, they claim it was impossible for their model which solved NS to have seen any user data since July 3rd, including codex chats.

 

[tomBitonti]

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.

What I've found, having taken graduate level courses in mathematical logic, and having spent decades using analytical methods when developing software, is that analytical thinking, while hard for most people, is easy if you are skilled in a particular analytic framework, and is very amenable to mechanistic steps. People don't care to carry through enough detail, and are rather not interested in mechanical reasoning steps. Analysis is not the forte of people.
Thanks!
TomB

 

[tomBitonti]

Something that strikes be, in relation to the posted description, and the proof which is attached to it, is how unelegant the solution is, and how unrevealing it is of the mathematical principles which one should take from the proof.

(For reference: https://openai.com/index/navier-stokes-solution/)

That is, an approach is outlined, and there follow a great many steps of carrying through to prove that the steps, when examined in detail, maintain desired bounds. This feels to me a mathematical rendering of an outline, very similar to a detailed graphical rendering that started with a wireframe model.

Put another way, the work done to incrementally build an understanding of the problem, with gathered intermediate results and insights, is missing, and is very important if the result is to be used to amplify human understanding.

TomB

 

[FrogReaver]

What I've found, having taken graduate level courses in mathematical logic, and having spent decades using analytical methods when developing software, is that analytical thinking, while hard for most people, is easy if you are skilled in a particular analytic framework, and is very amenable to mechanistic steps. People don't care to carry through enough detail, and are rather not interested in mechanical reasoning steps. Analysis is not the forte of people.
Thanks!
TomB

I don’t disagree with the sentiment but…

Analysis is a particular field within mathematics. It doesn’t map directly onto analytical thinking any more than another mathematical discipline does. Mathematical analysis has its own specific subject matter and methods. The Navier Stokes problem, for example, is a problem in mathematical analysis, not simply an example of “analytical thinking.”

 

[FrogReaver]

Link

In the same, they claim it was impossible for their model which solved NS to have seen any user data since July 3rd, including codex chats.

I’m slightly skeptical. There’s always the context window. Now, maybe those conversations simply weren’t included in the model’s context or otherwise made available to the agents after July 3rd, but the capability to provide that information is there. And we also don’t have the time frame for when the important insights originated.

I really wish that when companies made claims like this that they would elaborate on the evidence supporting them, especially when the claim is essentially that a particular category of data was not available to a system that otherwise had the technical capability to access it.

 

[The Firebird]

What I've found, having taken graduate level courses in mathematical logic, and having spent decades using analytical methods when developing software, is that analytical thinking, while hard for most people, is easy if you are skilled in a particular analytic framework, and is very amenable to mechanistic steps. People don't care to carry through enough detail, and are rather not interested in mechanical reasoning steps. Analysis is not the forte of people.
Thanks!
TomB

Agree with this. Ime LLMs are better than humans at analyzing routine scientific datasets, just because they implement the details that most humans would reject as unimportant or not worth the time. They are not yet as good at defining and orchestrating a scientific program.

 

[The Firebird]

I’m slightly skeptical. There’s always the context window. Now, maybe those conversations simply weren’t included in the model’s context or otherwise made available to the agents after July 3rd, but the capability to provide that information is there. And we also don’t have the time frame for when the important insights originated.

I really wish that when companies made claims like this that they would elaborate on the evidence supporting them, especially when the claim is essentially that a particular category of data was not available to a system that otherwise had the technical capability to access it.

My read is the training data for this model cut off on July 3rd (pre training? RL?) and it hasn't seen new data since.