Summer of Math Exposition

From Chess and Go to Olympiad-level math, DeepMind’s AI has tackled some of the hardest thinking problems known to humans. But here’s the surprising part: models like AlphaZero and AlphaProof don’t actually “think” like us — they turn these challenges into search problems. In this video, we explore how Go can be represented as a massive search tree with more possibilities than atoms in the universe, and how AlphaZero uses Monte Carlo Tree Search combined with neural networks to master the game. We’ll then look at AlphaProof, which takes a simpler but smarter approach to searching through proof steps, enabling it to solve Olympiad-level math problems and even reach a silver medal standard at the IMO. By the end, you’ll see how DeepMind reframed thinking itself into search — and why that’s such a powerful idea for the future of AI.