@rarelyquinn imo pursuing consistency is most productive when the model is important enough for the effort, sufficiently parameterized, and has access to data or ability to collect data to fill in unknowns
More notes on nearby ideas
Tweet · Oct 14, 2023consistency is overrated we're all continually building & revising our world model, a lossy abstraction on the complexities of the world, finding…Reply · Oct 8, 2023@emnode oh idk, consistency is overratedReply · Aug 18, 2023@gdb @yacineMTB 🤔 I do still feel like I'm missing intuition wrt model parameters & etc ... there's so much churn w AI I don't feel competentQuote tweet · Jun 12, 2024💯 this whole thread- truly open models (w training data) are so valuable for open research & understanding AI