OpenAI researcher proposes Kolmogorov Appendix for math papers
Sebastien Bubeck proposed that math papers include a Kolmogorov Appendix listing the minimal prompts needed for an AI to rediscover the central proof. The appendix would serve as an explicit record of novelty. Csaba Szepesvari responded that the value of such concrete guidance is path-dependent and applies equally to papers written by humans or by AI systems, centering the exchange on prompt-based reconstructions for greater transparency and reproducibility.
#50Sebastien Bubeck@SEBASTIENBUBECKMaybe math papers could include a "Kolmogorov Appendix" with the minimal prompt (or sequence of prompts) needed for an AI to rediscover the proof? With proper rules this could lead to nice explainable blueprint of the paper novelty! (Could have variants for low/high thinking too)
Like "when trying to solve the problem you will probably want to do X but that would be a deadend because of XYZ so instead do Y"
Sebastien Bubeck@SebastienBubeckMaybe math papers could include a "Kolmogorov Appendix" with the minimal prompt (or sequence of prompts) needed for an AI to rediscover the proof? With proper rules this could lead to nice explainable blueprint of the paper novelty! (Could have variants for low/high thinking too)
@SebastienBubeck this is quite path dependent. but yes, I often want just this regardless of whether the paper is written by human or AI
Sebastien Bubeck@SebastienBubeckLike "when trying to solve the problem you will probably want to do X but that would be a deadend because of XYZ so instead do Y"