Giuseppe Prencipe
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Human Interventions as Labels: Localizing, Steering, and (Sometimes) Editing the Computations Behind LLM Agent Failures
According to the abstract excerpt, the authors treat a human stepping in to correct an agent as a label pointing at the failing internal computation. In a controlled setting about needless clarifying questions, they report locating the decision inside models and flipping many over-asks.