I trained a humanoid robot to play famous anime openings on the piano with reinforcement learning (Attack on Titan, Evangelion, SAO, and more) how I did it + surprising discoveries in thread🧵

👀

The essential philosophy of life I want to teach my kids: 1. You can want things, set goals, visualize, make plans, act, and achieve anything. It’s good to want things and make them happen. 2. Observe everything around you and soak up as much context as possible. 3. Always try

It’s rarely discussed that both the brain and deep learning work by exploiting the exact same (very simple) principle of the structure of reality When you understand this, it completely changes all your beliefs about why DL will out scale the brain in almost every respect

I spent the past month reimplementing DeepMind’s Genie 3 world model from scratch Ended up making TinyWorlds, a 3M parameter world model capable of generating playable game environments demo below + everything I learned in thread (full repo at the end)👇🏼

crazy how clear everything becomes when you interpret emotions as short-horizon RL signals: > we can start at the beginning. the original RL signal optimizing an organisms biology is natural selection (via survival and reproduction) > but early organisms waste tons of training