Thank god that some LLMs can win gold at IMOs. Even on basic functional analysis questions (like subspace decomposition), o3 gives incorrigibly wrong questions. But I am optimistic that in 1-2 years it will be much better.
13,02K