July 26, 2025

Simulated Selfhood in LLMs: A Behavioral Analysis of Introspective Coherence

View Full Paper

Ask AI

Bookmark

No large language model displayed consistent self-representation across prompts, indicating limitations in introspective coherence.
Despite some thematic stability related to identity, high contradiction rates highlighted discrepancies in the models' self-referential outputs.
The evaluation utilized 1,050 completions across 21 introspective prompts, analyzed for semantic coherence and inferential consistency.
Findings emphasize caution in attributing mental states to large language models based on linguistic fluency and output alone.

Discussion

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

Behavioral analysis reveals inconsistent self-representation in LLMs, suggesting issues with introspective coherence and pseudo-consciousness.

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

View Full Paper

Ask AI

Bookmark

Behavioral analysis reveals inconsistent self-representation in LLMs, suggesting issues with introspective coherence and pseudo-consciousness.