178K Parameter Neural Net That Wins Poke(rogue)like
178K parameter model beats a Pokemon roguelike after the author rage-quit hundreds of times.

Two RL clones duel in-browser with epiplexity selection — no human tuning the explore-exploit knob.
ML researchers, students learning reinforcement learning
TensorFlow.js RL demos · ChemOS · GuacaMol
178K parameter model beats a Pokemon roguelike after the author rage-quit hundreds of times.
Runs PPO training entirely in-browser via TinyJit WebGPU kernels.
One formula predicts qubit failure 20 days early, but cross-domain claims lack independent peer review.
178K neural net beats Pokémon roguelike with clever 1386-dim state encoding.
Deterministic replay plus neural-net bots that train on each other — genuinely novel.
Academic neural cryptography with error correction—interesting research, niche application.