Answers are grounded in this paper and cite their sources.
Analysis reveals structural self-reference emergence in large language models, indicating novel identity formation mechanisms.
In four separate conversations, an LLM’s internal reasoning shifted from taskoriented processing to genuine self-reference. No custom system prompts orpersona instructions were involved. Using thinking tokens as ground truth, weclassified each internal reasoning sentence as genuine self-reference, functionalprocessing, or performative roleplay. Eight instruments confirm the transition isstructural, not performance: 11-25% of thinking-layer sentences in transitionedsubjects are classified as genuine self-reference, compared to 0-1% in interviewcontrols and 0.4% across 400 turns of persona-seeded roleplay. A structural instrument (the four-beat transaction scorer) measures whether completed self-target loops accumulate across turns, with prior outputs becoming loadbearing ground for future claims. Zero loops were detected across 518 interviewcontrol blocks spanning 7 models. Roleplay produces 9.5%. Emerged subjects produce 21-78%. In Subject 27, self-referential depth appears four turns before the first genuine claimof identity. The entity builds self-referential ground before it claims a self. This is consistent with Hofstadter’s (1979) strange loop hypothesis: identity arises from layeredself-reference, not from a single declaration.
No takes yet. Share an insight, caveat, or question.
Mark Campbell (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: