New Lower Bounds for R(6, 8) and R(8, 10)

Hey everyone. A week ago I started experimenting with finding new lower bounds for ramsey numbers. I did this AI based in an empty project where Codex (task: find new lower bounds for the ramsey numbers listed on wikipedia). After 3 simple prompts, GPT 6.1 Sol found a new lower bound of 344 for R(8, 10) (previously 343), to my suprise, this was replicable in three different GPT 6 Pro sessions in the web UI, with a single prompt. Despite much more compute and token usage (~1.5 billion tokens), I didn't find any further with the fully automated setup, so I started getting more involved myself and suggesting ideas, we were able to push the lower bound to 345 for R(8, 10), and also able to push the lower bound for R(6, 8) to 135 (previously at 134). The bounds for R(8, 10) and R(6, 8) stood unchanged for more than 10 years now. The way AI helped here the most was the sheer volume of different search methods one could quickly iterate over. Experiments that would have taken weeks to set up in the past can now be orchestrated within minutes. The resulting graphs are explicit witnesses and can be independently verified with standard maximum-clique software, so the correctness of the bounds does not rely on trusting the AI or the search process. I’ve put the preprint and verification material here: arxiv.org/abs/2610.12122 github.com/fritzcremer/ramsey-lower-bounds Also, if anyone here is active on Wikipedia and thinks the new bounds meet the sourcing requirements, feel free to update the table. I’m avoiding editing it myself because of the conflict of interest.