@ESYudkowsky seriously tho, I don't find this all that surprising and had been assuming I could do something similar to compress chats into summaries to get to approx the same spot in latent space w fewer tokens
More notes on nearby ideas
Reply · Nov 8, 2024@DaveShapi is intelligence not compression?Reply · Jan 24, 2022@ionlazer Feel like once we're uploaded we can classify NPCs quantitatively by how much they can be compressedReply · Apr 13, 2023hah! sometimes I think 3.5 is easier to coax into a "botsonality" that's less... umm... fake?Reply · Aug 31, 2023@max_paperclips yeah, it's a struggle to figure out how to explain embeddings and GPT functions fast enough before ppl's attention span blinks out