harness² from @GoogleResearch offers techniques for RSI on the harness, with a fixed model they call it "harnessing", but you might already be familiar w/ this process! it's also called "hill climbing" or "continual learning" more in 🧵 !

→ View original post on X — @langchain