The LLM got closer than ever before to cracking the cap set problem, which has perplexed mathematicians for decades.