In August 2026, OpenAI announced solutions to ten major mathematical and theoretical computer science problems, including the first construction of a non-sofic group and a proof that the multicolour Ramsey number $R(3, 3, \dots, 3)$ grows superexponentially. While these results are significant, the author observes that LLMs do not yet outperform humans in all mathematical aspects. Instead, LLMs appear to excel at finding "examples" or "counterexamples"—existence statements that disprove long-standing conjectures or demonstrate non-trivial properties—rather than proving complex universal theorems. The author theorizes that this is due to a distinct "LLM style" of mathematics characterized by two strengths: vast mathematical knowledge and high-speed computational iteration. While humans often rely on "surprising" or "conceptual" leaps, LLMs likely succeed through a probabilistic "try-lots-of-things-until-you-get-lucky" approach. This method aligns with several mathematical strategies for finding examples, such as testing off-the-shelf objects, building constructions from basic components, utilizing metavariables to describe generalized classes, or attempting to prove the negation of a statement to identify necessary intermediate lemmas. Ultimately, the author suggests LLMs are highly efficient at solving existence problems where the solution can be reached through iterative, non-novel, yet rapid trial and error.