CLM vs Jev evaluation
A comparison of CLM and Jev as game controllers, on bounded agent decisions, and on 1,253 public function-selection cases. Explore the methods, results, and individual answers.
Experiments
I build and run experiments to see how an idea works in practice. Some are demos you can try; others show what I tested and what happened.
-
-
Cross-tab WebGPU coordination
A browser experiment that keeps one WebGPU language model in memory when several tabs are open. Web Locks choose a leader tab; BroadcastChannel carries prompts and streamed replies to the others.
-
Text Watermark Demo
A browser-local demo of statistical text watermarking. Generate text with a small WebGPU language model, inspect the watermark signal, then edit the text to see what changes.