This market resolves to the event which happens first:
Event 1: A quantum computer is capable of factoring RSA-2048 (the number has to be publicly factored, and the demonstration has to be verified).
Event 2: An AI model is capable of autonomously generating annals-quality math papers for under 100K$, as specified in this market:
https://manifold.markets/TamayBesiroglu/will-ai-be-capable-of-producing-ann
If neither of the events happen by January 1, 2040, the market resolves to the third option.
Update 2026-08-03 (PST) (AI summary of creator comment): - Although the linked market by Tamay Besiroglu resolved early due to a concession, this market will remain open until the criteria for one of the events fully holds.
Update 2026-09-11 (PST) (AI summary of creator comment): - To satisfy Event 2, the AI must be capable of autonomously writing readable, well-explained papers.
Generating correct results, raw notes, or formal proofs (like Lean) that require human translation or rewriting to be understandable will not count for resolution.
1,000Market update:
The models are obviously very capable - if I had 100K$ to spare, I could probably let Astra run autonomously and get an annals-level result, accompanied by a Lean proof. OpenAI's internal model is even better.
However, this is still not an annals-quality paper, because AI's writing is sloppy and it can't explain its results well.
This is not a nitpick; I am using these models for research, and the notes that Astra and Fable generate for me are basically unreadable. Most of my effort goes into understanding their results and rewriting them in readable language. I expect a true autonomous mathematician to be able to explain their results in a way other people can understand.
Thus, I am not resolving this market for now. I would like to hear your thoughts on this - am I wrong? is there a way to make these models generate good writing that I'm just missing out on?