This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
Tim Kellogg
timkellogg.me
did:plc:ckaz32jwl6t2cno6fmuw2nhn
Self-Improving Transformers
They found that you can train LLMs on their own outputs by
1. generating *slightly harder* problems each time
2. filtering low quality via majority voting
Is this the singularity? Maybe, but I think it might just be benchmark saturation.
https://timkellogg.me/blog/2025/02/12/recursive-improvement
2025-02-13T03:39:40.516Z