Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2512.12411 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911475187056640 |
|---|---|
| author | Hahami, Ely Sinha, Ishaan Jain, Lavik Kaplan, Josh Hahami, Jon |
| author_facet | Hahami, Ely Sinha, Ishaan Jain, Lavik Kaplan, Josh Hahami, Jon |
| contents | 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 demonstrate that LLMs can compute meaningful functions over perturbations to their internal states, establishing introspection as a real but layer-dependent phenomenon that merits further investigation. Our code is open-sourced here: https://github.com/elyhahami18/llama-introspection-new |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_12411 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Detecting the Disturbance: A Nuanced View of Introspective Abilities in LLMs Hahami, Ely Sinha, Ishaan Jain, Lavik Kaplan, Josh Hahami, Jon Artificial Intelligence 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 demonstrate that LLMs can compute meaningful functions over perturbations to their internal states, establishing introspection as a real but layer-dependent phenomenon that merits further investigation. Our code is open-sourced here: https://github.com/elyhahami18/llama-introspection-new |
| title | Detecting the Disturbance: A Nuanced View of Introspective Abilities in LLMs |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2512.12411 |