Can “consciousness” be observed from large language model (LLM) internal states? Dissecting LLM representations obtained from Theory of Mind test with Integrated Information Theory and Span Representation analysis

Jingkai Li

Abstract: Highlights•We explored LLM representations through the lens of consciousness studies.•We used a triangulated method to observe ”consciousness” in LLM representations.•Findings stem from multiple LLM layers, diverse stimuli spans, and various ToM tasks.•No strong evidence of ”consciousness” but intriguing patterns via spatio-permutation.

External IDs:doi:10.1016/j.nlp.2025.100163