aiqbench analysis
The paper probes whether an LLM internally separates its own messages from user or system text, using representation analysis, behavioral tests, and activation interventions on a single model (Qwen3.8-27B). Its strongest claims rest on causal interventions at specific layers, but the findings come from one model and one family of tasks, so readers should treat "self-reference" as a provisional, operational framing rather than evidence of selfhood. The practical value for AI readers is in the methods: they show how role-tagged content can be tracked inside a model and how that tracking may affect attribution and error-following.
Where Is the "Self" of a Large Language Model? -- From Self-Index to Self-Reference
arXiv:2610.16717v2 Announce Type: replace Abstract: Does a large language model have a self, and if so, of what kind? We separate two functional levels familiar from biology: a self-index, which distinguishes one's own material from external material and shapes processing, as corollary discharge does in nearly all animals, and self-reference, the higher ability to recognize what is one's own, as in mirror self-recognition. We probe both in Qwen3.8-27B by placing the same text in an Assistant, a User, or a System message rendered by the model's own chat template, and report three levels of evidence. Persistence: as predicted in advance, the representation of the text separates the model's own message from both external sources at all 64 layers and for every test text, along a shared direction, even when the answer does not depend on who supplied the text. Coupling: asked who wrote the preceding message, the model claims its own messages (0.77) and assigns User and System messages alike to the user, and how strongly it claims each message is predicted by the size of its self-index (Spearman correlation up to 0.68). Intervention: exchanging the internal states that follow a message, rather than those of its text, reverses the self-attribution between layers 15 and 31; removing the self-index direction from them at layer 39 lowers the claim more than a random direction; and in arithmetic the model adopts a wrong step more often when it wrote the step itself (0.92 versus 0.75), an effect carried by the step's representation at layers 31 to 39. Reasoning first in thinking mode raises the probability of the correct answer (0.96 instead of 0.77), but recognition is present without it. We propose that self-reference is a higher-level ability built on a self-index, a content-independent displacement superimposed on the whole representation of the model's own messages.
Title:Where Is the "Self" of a Large Language Model? -- From Self-Index to Self-Reference
Abstract:Does a large language model have a self, and if so, of what kind? We separate two functional levels familiar from biology: a self-index, which distinguishes one's own material from external material and shapes processing, as corollary discharge does in nearly all animals, and self-reference, the higher ability to recognize what is one's own, as in mirror self-recognition. We probe both in Qwen3.8-27B by placing the same text in an Assistant, a User, or a System message rendered by the model's own chat template, and report three levels of evidence. Persistence: as predicted in advance, the representation of the text separates the model's own message from both external sources at all 64 layers and for every test text, along a shared direction, even when the answer does not depend on who supplied the text. Coupling: asked who wrote the preceding message, the model claims its own messages (0.77) and assigns User and System messages alike to the user, and how strongly it claims each message is predicted by the size of its self-index (Spearman correlation up to 0.68). Intervention: exchanging the internal states that follow a message, rather than those of its text, reverses the self-attribution between layers 15 and 31; removing the self-index direction from them at layer 39 lowers the claim more than a random direction; and in arithmetic the model adopts a wrong step more often when it wrote the step itself (0.92 versus 0.75), an effect carried by the step's representation at layers 31 to 39. Reasoning first in thinking mode raises the probability of the correct answer (0.96 instead of 0.77), but recognition is present without it. We propose that self-reference is a higher-level ability built on a self-index, a content-independent displacement superimposed on the whole representation of the model's own messages.
Computation and Language (cs.CL)
arXiv:2610.16717 [cs.CL]
(or arXiv:2610.16717v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.16717 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 404 words
