Jacob Andreas
2 indexed papers
Publications per year
Top categories
Frequent co-authors
Research Timeline
The paper proposes a method to approximate the behavior of attention heads in transformer language models with executable programs.
This paper explores when language models (LMs) generate faithful introspection during explanation training, finding that LMs can produce more faithful explanations from earlier checkpoints or similar models, even when their behaviors shift.
Papers
Introspective Coupling: Self-Explanation Training Tracks Behavioral Change Despite Fixed Supervision
This paper explores when language models (LMs) generate faithful introspection during explanation training, finding that LMs can produce more faithful explanations from earlier checkpoints or similar…