Internship over, time to go finish my degree.
Had a great time at OpenAI. I am now certain 2027 will see the dawn of mathematical superintelligence and it excites me to eventually likely see a typical century’s worth of mathematical progress happen in the span of a few months.
4) Shifting focus to the long gaps result, I was amazed by how elementary the proof is and how it seemingly slipped by all analytic number theorists (the Lean formalisation is only 4.5k LoC!) and is SoTA, improving further beyond the result found by Sol on Erdős problem 4.
1/7 Okay, I’ve spent my lunch digesting the paper more and I think Claude didn’t make much of the contribution here. A lot of the key work was of BGSTB24/25 which I was familiar with when these papers released and were pretty big deals.