This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
mr. TIM
timkellogg.me
did:plc:ckaz32jwl6t2cno6fmuw2nhn
this is pretty cool — @goodfire.bsky.social is using internal signals, mech interp, as reward signals in RL training
they replicated the J-Space paper on GLM-5.2, trained a Qwen reward model to identify hallucination in J-Space signals, then reduced hallucinations 37%
https://x.com/goodfireai/status/2077073005088501780?s=46&t=ftkDjGBpGPr2-yTN2CCUYg
2026-07-15T02:26:23.073Z