@cwebber Does not take a medal of any sort to see that no-one is cool about LLMs producing results faster than humans can follow. Also: your paying customers/boss can check 2 hours of your work in 2 minutes with an LLM, so you are inevitably driven to cross-checking your own work with LLMs first.
So we are entering a new paradigm of overload & exhaustion for 100% intellectual professions. The harm done is a price we pay for any benefits. But asking it to stop is c..ute like Canute.

@cwebber Tao was previously cheerleading the use of AI to find proofs but I guess it's good he's figured out the scam?

@cwebber In other words, cheating and plagiarism do not bring anything to the table.

@cwebber well, I think AI businesses addressing science is good. AlphaFold is a great good after all, speaking from the personal expertise. Yet this is good too:

>In many fields and activities, years of training have traditionally served not only to produce a final answer or product, but also to develop understanding and the ability to formulate new questions and ideas.

Personally, I would like to see the graveyard - a list of problems where the solutions did not converge.

@cwebber I wish they put a date to the letter.

@cwebber I love the prose of this very much, it captures the problem extremely well, but I utterly despise the concession at the end.

"We fully recognise the danger of this on our field, but we should allow ourselves a little, as a treat!"
We've seen over the past few years just how much that doesn't work & only lets those that are known to take a mile whenever given an inch to do exactly that.

Now, on the other hand, the guys over at AHM are people to watch.
https://www.ahmath.org/

@cwebber Stupidifying humanity is what this type of LLM usage is achieving.

And it’s all due to will to power, not intellect.

@cwebber @AndrewHenry “you’re plagiarizing out work and if you don’t stop, we’ll eff you up” would be powerful. This is conceding ground from the get go by giving into “LLMs can do math now” narrative and the alignment concept which is specifically designed to muddy waters regarding discussion about generative ai regulation.

I’m honestly disappointed.

@cwebber I'm surprised that the letter says nothing of the fact that OpenAI effectively scooped Navier-Stokes by hiring an entire team of mathematics PhDs to steal^W learn from mathematicians that were busy pouring their mental diaries on the problem into Codex. Has that story changed in the past day or two?

Standing on the shoulders of human beings that boxed a problem in sufficiently enough to allow brute-forcing a final proof in a formal proof language like Lean isn't nearly as impressive to me as everyone else seems to think it is 🤷