3 ms·
Latent Programming Horizons in Coding Agents
- ahana_537 4mo agogood to read
- sparsesignal 4mo agoNice paper. They predict the outcome of edits up to ~25 steps before the agent makes them. Decodable doesn't mean causal, as the authors note, but a cheap probe that flags doomed trajectories early could save a lot of wasted agent compute. We clearly still have a lot to learn about what these models represent internally.
- ahana_537 4mo ago[dead]