Frontier Research
Models, research, safety and the ideas pushing the frontier of AI
AI Reasoning
How an AI won gold at the world's hardest maths competition
Google DeepMind's system solved problems that defeat almost every human who attempts them, and it did so in a way that cannot bluff: every line of its reasoning was checked by a machine that only accepts proof
18 min read
Computer Vision
The trick that let neural networks finally go deep
Piling more layers onto a neural network made it worse, not better, until a Microsoft team added a shortcut that let each layer learn only what the ones before it missed
7 min read
Reinforcement Learning
The match that showed machines could master Go a decade early
Go's search space is so vast that brute force is hopeless, so DeepMind built a program that learned the game's intuition from human play and its own, then beat a champion nobody expected it to touch for years
9 min read