Replying to
@tao@mathstodon.xyz
@tao@mathstodon.xyz we are at the start of a New Enlightenment. LLMs make it easy to get empirical results quickly, form theories, then verify them with predictive tests.
The issue is we can verify theories before we understand them. Personally: I have an unpublished formal theory for my own public work on agent reliability, thanks to AI. But it is at the edge of my understanding.
I'm handing this by posting one concept at a time. Each one builds my understanding. Once done, they make the paper I will publish.