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

Steering identity with WeirdChat: contrastive directions induce disidentification but not introspection

Ian Rios-Sialer · Team Unruly Abstractions

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

Read the report

Report: Steering identity with WeirdChat: contrastive directions induce disidentification but not introspection

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

Can a model be steered to disidentify with being an AI assistant, and can it detect that steering? We test both on Qwen3.6-27B, using two WeirdChat behaviors as target states: claiming a physical body and denying being an AI. We compare two steering methods. Direct contrastive steering learns each behavior direction as the difference in mean activations between matched exhibiting and non-exhibiting replies; the directions separate held-out replies for both behaviors, and steering turns the body behavior on in fluent text, as confirmed by a judge using WeirdChat's own rubric. A regulator choosing signed strengths on word-list J-space directions steers nothing: a norm-matched decoy moves the readout more than the target. A forced-choice letter probe confabulates at zero injection, reading the conversation's own behavior as an injection. Free-text answers at zero injection are "none" in all but two trials; under injection, they fill with the injected concept's vocabulary instead of naming it. Introspection reports explicit context, not internal steering.

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. As explained clearly in the introduction, this project asks two questions: "Can we steer a model to disidentify with being an AI assistant?" and "Can a model detect such steering?" The answers it reports are, respectively, "Yes, with contrastive steering, but not a J-space-based approach" and "No". Unfortunately, I found the write-up of the methods and results hard enough to follow that I'm not well-positioned to critique the details.

  2. Interesting approach with an interesting limit - embodiment succeeding while denying being an AI failed is at least suggestive about how easily steered different aspect of identify and experience might be. Worth following up and expanding.

Cite this project

@misc{riossialer2026steering,
  title = {{Steering identity with WeirdChat: contrastive directions induce disidentification but not introspection}},
  author = {Ian Rios-Sialer},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/steering-identity-with-weirdchat-contrastive-directions-induce-disidentification-but-not-introspection-yt9t}},
  url = {https://apartresearch.com/sprints/projects/steering-identity-with-weirdchat-contrastive-directions-induce-disidentification-but-not-introspection-yt9t}
}

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