August 15, 2025Open Access

Simulated Selfhood in LLMs: A Behavioral Analysis of Introspective Coherence

View Full Paper

Ask AI

Bookmark

No model maintained consistent self-representation despite high thematic stability in some prompts.
Five LLMs produced 1,050 outputs analyzed across various behavioral dimensions including semantic coherence.
Observation approach utilized structured introspective prompts to evaluate models' outputs' internal coherence.
Caution is advised in inferring mental states from language fluency alone, highlighting LLM interpretability challenges.

Discussion

No takes yet. Share an insight, caveat, or question.

Behavioral analysis examines self-representation in LLMs using introspective prompts, indicating complexities in interpretability.

Jose Augusto de Lima Prestes (2025) studied this question.

View Full Paper

Ask AI

Bookmark