Watch a neural net discover molecules by arguing with itself
Two RL clones duel in-browser with epiplexity selection — no human tuning the explore-exploit knob.
Training-free, token-free topological collapse engine. No training, no tokens, no neural net. Runs offline. AGPLv3.
Persistent homology across three domains with honest failure reporting.
Data scientists and researchers working with signal analysis
giotto-tda · Ripser · scikit-tda
Two RL clones duel in-browser with epiplexity selection — no human tuning the explore-exploit knob.
178K parameter model beats a Pokemon roguelike after the author rage-quit hundreds of times.
258 parameters controlling 2.2M connections with 8,384:1 compression is wild.
MLP learns your nav patterns—clever, but Telescope and fzf already solve file finding.
One formula predicts qubit failure 20 days early, but cross-domain claims lack independent peer review.
Recovers Newton's gravity from raw signal prediction using a bandwidth-limited GRU.