Jonathan Stray @jonathanstray.com · 22d There's a long list of content-based signals we might want to rank on. We are limited by the classifiers we actually have. BUT we are working on LLM scoring with arbitrary prompts! The challenge is performance, but we have a cunning plan docs.google.com/document/d/1...
3 likes 2 replies
?
Replies
This is ofc very efficient, but do you have stats on cross validation with this method? How well does this generalize?
Performance is gonna be a huge issue if you try to analyze every post right? So that's a reason I mentioned topic trust maps. You can use who follows who & who agrees with who about what as a proxy to figure out which posts are worth feeding to your LLM.