interpretability
1 statements across 1 episodes · 0 bullish · 0 bearish · 1 people on the record · first statement Oct 23, 2025 by Julian Schrittwieser · across every show →
Everything said about interpretability, oldest first
Oct 23, 2025
Using chain-of-thought as an RL reward destroys model interpretability
“If you're not careful with RL, you can make interpretability harder. For example, one Common thing with modern models is they do reasoning with the chain of thought. You could look at the chain of thoughts to, you know, see what are the model internal thoughts…”