✳flâneur — a map of the web's best reading

Detecting the Disturbance: A Nuanced View of Introspective Abilities in LLMs

arxiv.org · 6,754 words · saved by 1 readers

Can large language models introspect, that is, accurately detect perturbations to their own internal states? We systematically investigate this question using activation steering in Meta-Llama-3.1-8B-Instruct. First, we show that the binary detection paradigm used in prior work conflates introspection with a methodological artifact: apparent detection accuracy is entirely explained by global logit shifts that bias models toward affirmative responses regardless of question content. However, on tasks requiring differential sensitivity, we find robust evidence for partial introspection: models localize which of 10 sentences received an injection at up to 88% accuracy (vs. 10% chance) and discriminate relative injection strengths at 83% accuracy (vs. 50% chance). These capabilities are confined to early-layer injections and collapse to chance thereafter—a pattern we explain mechanistically through attention-based signal routing and residual stream recovery dynamics. Our findings demonstrat

Detecting the Disturbance: A Nuanced View of Introspective Abilities in LLMs Ely Hahami Ishaan Sinha Lavik Jain Josh Kaplan Jon Hahami Abstract Can large language models introspect, that is, accurately detect perturbations to their own internal states? We systematically investigate this question using activation steering in Meta-Llama-3.1-8B-Instruct. First, we show that the binary detection paradigm used in prior work conflates introspection with a methodological artifact: apparent detection accuracy is entirely explained by global logit shifts that bias models toward affirmative responses re

Explore this link on the map →

related reading