This may have been "solved" using anonomized data from scientist already testing the same theory. May want to hold resolution til a confirmation of who is credited with solving the problem.

A likely scenario: mathematicians worked with and prompted models to progress through to a solution, where the prompting aided progress. This is right in the gray zone of your resolution criteria described in your comments below.

Does it have to be certified by Clay?

@LukaszWiklendt Looks like it can be resolved to YES: AI may have just solved a million-dollar math problem. The field will never be the same

AI may have just solved a million-dollar math problem. The field will never be the same

A mathematician compared the feat to the history-making chess competition in which IBM’s Deep Blue computer beat Garry Kasparov in 1997

@SurvivalBias likely YES, but let's wait for some more thorough review of the claims and ideally a statement from millennium prize.

@SurvivalBias surely it depends on the resolution criteria (which are missing) and the details of how the solution came about. How much human contribution does it take to not be an AI solution?

@SurvivalBias this cannot be resolved yet. the proof has not been checked by experts. yes, there is lean code, but it has not been checked for semantic misalignment. it's also unclear if this form of N-S is enough for the millenium prize to be awarded (though it probably is)

what does "solved by AI" mean? does an AI being involved in any part of the process count or does it have to be the majority contributor?

@Balasar to me, it's one of those where the people presenting the proof say "without AI, we wouldn't have been able to solve it".

https://youtu.be/cru2bkqwSYk?si=VVOcdZTEoTGusCMK&t=1012

https://youtu.be/E6hCFDfkijU?si=vYi-hxSvlQl-8-NZ&t=1079

https://x.com/apples_jimmy/status/1938024622177604082

Note: text says "end of 2028" but market closes end of 2027.

There's a wide gray zone between "solved by an algorithm all on its own" and "a mathematician used computer to help with an intermediate step in the solution" - where do you draw the line on this? The wording of the title suggests you'd only count if an algorithm solves the problem all on its own - is that right? Would that still count if a special algorithm was purpose built for the problem at hand or would that count as "human with help from an algorithm"?

@MartinModrak Something along the lines of the 4-colour theorem would not count. AlphaTensor would count since it is not just an enumeration of all combinations, it required learning a set of parameters that somehow captured the essence of the problem and then solved it. A prompt simply asking the question in terms that an undergraduate maths student could understand and the AI producing the correct solution as a response would very much count.

@LukaszWiklendt You named some things that would count and some things that wouldn't, but not how we could evaluate a novel scenario.