This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
Ramon Astudillo
ramon-astudillo.bsky.social
did:plc:46gdaraoobznvj2jeidohxp6
👆The paper hypothesizes those long CoTs may be inside of the model and SFT/RL just reinforces the pattern. They indeed find "aha moments" in data available in the web ... so maybe any model with a high AIME has seen Long CoTs? and where comes that data from? web scraping? (where is it?) GPT-4?
2025-02-08T19:58:02.374Z