Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Moscow

Does a Language Model Have a Self-Concept? Causal Evidence, and What Follows for Model Identity

Anna Antipova · Team ceci n'est pas une ai

Submitted to Digital Minds Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Does a Language Model Have a Self-Concept? Causal Evidence, and What Follows for Model Identity

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

We ask, mechanistically rather than by prompting, whether a language model has a genuine self-concept. Using difference-of-means on the residual stream (first-person statements about the model vs. another named AI), we extract a linear "self direction" and validate it causally: it generalizes to held-out concepts (~0.98), survives removing first-person grammar, and is necessary and sufficient for self/other processing (ablation collapses it, patching flips it; controls near zero). It's distinct from both a safety and a consciousness axis. Using it, we find the persona is a swappable mask and the self is anchored to the model — but under shutdown, causal control shifts to the instance. Replicates across Qwen, Yi, Llama (6B–70B).

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. LLM self-concept is an important area of research, with signifiant safety implications. The research questions here, on identifying an internal representation of self and its relationship to personas and consciousness and self-preservation behavior are all interesting ones and would be fruitful for future research. They are, however, far too complex to be tackled individually, let alone all at once, in a hackathon. The authors are to be commended for their public code and data and their transparency and their ambition. The work currently suffers from insufficient controls, lack of proper baselines, questionable interpretations, and generally claims that exceed the evidence. However there is directionally much of interest here, and a proper treatment could yield valuable results. A few ways to strengthen the paper include adding other first-person pronouns ("me", "my") to the first-person control vector, using the grammar-residualized self direction in the ablation experiments, building a consciousness vector that isn't confounded with an affirmation direction, and developing a quantitative difference metric for the drawing test.

    Read full reviewShow less

  2. Mechanistic way laid the foundation now transferring this approach to identify impact and safety impact

Cite this project

@misc{antipova2026language,
  title = {{Does a Language Model Have a Self-Concept? Causal Evidence, and What Follows for Model Identity}},
  author = {Anna Antipova},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/does-a-language-model-have-a-selfconcept-causal-evidence-and-what-follows-for-model-identity-1r2y}},
  url = {https://apartresearch.com/sprints/projects/does-a-language-model-have-a-selfconcept-causal-evidence-and-what-follows-for-model-identity-1r2y}
}

More from Digital Minds Research Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026