Mechanistic Interpretability of Socio-Political Frames in Language Models
This work addresses the problem of interpreting how LLMs encode human concepts like socio-political frames, which is incremental as it applies existing mechanistic interpretability methods to a new domain.
The paper investigates how large language models generate and recognize socio-political frames, showing they can fluently produce and identify these frames in zero-shot settings, with specific dimensions in hidden representations correlating strongly with frame presence.
This paper explores the ability of large language models to generate and recognize deep cognitive frames, particularly in socio-political contexts. We demonstrate that LLMs are highly fluent in generating texts that evoke specific frames and can recognize these frames in zero-shot settings. Inspired by mechanistic interpretability research, we investigate the location of the `strict father' and `nurturing parent' frames within the model's hidden representation, identifying singular dimensions that correlate strongly with their presence. Our findings contribute to understanding how LLMs capture and express meaningful human concepts.