#autonomousresearch

Originally published on andrew.ooo — visit the original for any updates, code snippets that aged...

I let an autonomous agent run 100 ML experiments while I slept. 7 succeeded. Net result: 25% model improvement. Here's the setup.