--- format: html: title: "Jayrup Nakawala" --- JAYRUP NAKAWALA -------------------------------------------------- Independent AI-safety researcher I build and study systems where models act, decide, and sometimes misbehave — usually the small ones, because small models are all I can afford. You can train one overnight, take it apart, and check whether what you believe about it is actually true. My taste in research: - Small, testable experiments over big claims \ - Simple baselines that beat fancy methods \ - Things that shouldn't work but kind of do \ - Mistakes that are found and fixed, out in the open \ I write about all of it the same way I think about it: from first principles, in plain words, so anyone can follow along. If I can't explain something simply, I don't understand it well enough yet. I prefer tools that run locally, logs that can be inspected, and experiments that fail loudly. -------------------------------------------------- [Note] If you are on Windows use curl.exe -------------------------------------------------- Feel free to reach out!