I will spend some time attempting to use AI to assist me in solving open Erdos problems. I am an undergrad and have some math background. I did alright on the Putnam. If I do it, I will post publicly available proof.
1,000@traders Using GPT 5.5 Pro I found an interesting improvement to the lower bound exponent on Erdos problem 90 recently solved by OpenAI (the unit distance conjecture) that is currently still the highest lower bound. Obviously, this is nowhere close to actually solving a new problem, and the approach was simply following the path outlined by Sawin's paper. However, I wanted to note this because I was surprised by the capabilities of even the non-internal models I have access to. One of the improvements was not merely a using a set of different primes, but finding a potential modification to a lemma in Sawin's paper.
https://teorth.github.io/optimizationproblems/constants/84a.html
https://mathoverflow.net/questions/511514/what-is-the-unit-distance-exponent/511531#511531