bbzz

[Stephen Wolfram @stephen_wolfram]

Since all sorts of people have asked me, I decided to write down my current thoughts on "What's the Future for Pure Math Research in the Age of AI?"
https://t.co/TQxHFIirSG

[bbzz @bbzz1067029]

@KnightShore @davikrehalt @nihilsuncle AI won't solve everything, and credit assignment won't die. Although it solves NS, llm is just a software capable of automatically handling many known tasks. Much creative work remains for humanity to accomplish, and credit is there. Yet the academic environment is not prepared.

[bbzz @bbzz1067029]

@wtgowers Perhaps using AI can be viewed like "citing results from a paper": often, people use it by grasping the statement and the underlying intuition—trusting it based on its publication—without fully going through every detail of it. Then people can build new things modularly.

[bbzz @bbzz1067029]

@nihilsuncle Perhaps we should re-evaluate work based on the methods involved; experts can distinguish between problems that are easily solvable by AI and those that are not.(2/2)

[bbzz @bbzz1067029]

@nihilsuncle Yes, it's a problem. The upending of evaluation systems is a major challenge for math. While there remain many problems beyond the ability of AI, the academic system appears to be faltering. (1/2)

[bbzz @bbzz1067029]

@nihilsuncle Another issue is: why can you be certain that GPT-based or formal verification is reliable? Current accuracy rates for mathematical verification fall far short of the level achieved in integer arithmetic. Responsible individuals should be skeptical of all AI-generated content.