Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 16, 2026Turin, Italy

Whose Goals Does the Functional Welfare Axis Track?

Yasin Edin · Team Whose Welfare

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

Read the report

Report: Whose Goals Does the Functional Welfare Axis Track?

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

The "functional welfare axis" is a direction in a language model's activations that tracks how well things are going for it (Han et. al, 2026). It was validated using a stimulus where a user says "That's right" or "That's wrong" — which cannot separate the model succeeded from the model's user is pleased. We separated them: 24,000 activation captures crossing the model's actual answer correctness with a counterparty's task-irrelevant outcome and declared identity (human user / AI peer / uninvolved third party).

The axis is dominated by the model's own outcome (d = 2.75) but genuinely tracks the other's (d = 0.49), surviving three sentiment controls. That coupling is nearly flat across identities — an uninvolved stranger's fortunes move it ~90% as much as the user's — and arrives from pretraining unchanged, while post-training amplifies own-outcome tracking 2.4×. Welfare readings taken mid-conversation are therefore partly a readout of the conversation, but the contaminant is generic other-regard, not user-approval tracking.

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. This is a clean, well-scoped correction to a specific measurement claim in the welfare-axis literature. The core design — holding the counterparty's verdict fixed to ground truth while independently varying their unrelated outcome and declared identity — genuinely isolates something the original paper's "That's right/wrong" stimulus could not, and the finding (own-outcome dominates ~5.6:1 over other-outcome, but other-outcome is real, survives three sentiment controls, and is present even for an uninvolved third party) is a useful, quantified correction rather than a vague caveat. The honesty throughout is a strength: reporting a failed pre-registered gate, reporting layer-sweep exceptions rather than only the best layer, and reporting that the two "independent" valence axes disagree more than expected. The main thing that would make this stronger is acknowledged in the paper's own limitations — a false-verdict 2×2 to fully separate own-outcome from being-told-you-succeeded, and a steering arm to move from correlational to causal. Both are clearly the right next experiments and are already flagged as such.

    Read full reviewShow less

Cite this project

@misc{edin2026whose,
  title = {{Whose Goals Does the Functional Welfare Axis Track?}},
  author = {Yasin Edin},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/whose-goals-does-the-functional-welfare-axis-track-6ae5}},
  url = {https://apartresearch.com/sprints/projects/whose-goals-does-the-functional-welfare-axis-track-6ae5}
}

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