Skip to content

Topic

Chain-of-thought extraction

Harvesting the intermediate reasoning a model produces before its final answer, which vendors variously expose, summarise or encrypt because those traces are valuable training data.

Current clusters