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

Base Model Persona Inference can Predict Misalignment and Surface Agreement

Arush Tagade, Taslim Mahbub · Team Praxis Research

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

Read the report

Report: Base Model Persona Inference can Predict Misalignment and Surface Agreement

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

Language models have a consistent assistant persona that is the product of post-training but this assistant character is not well understood. We introduce base model persona inference that assigns a persona to a post-trained LLM response and show that continuations sampled from this persona can: 1) predict broad misalignment caused by harmful responses and 2) provide consistent frames of analysis between revealed and stated scenario preferences. We expect base model persona inference to become a strong tool in the toolkit for assessing model preferences and hidden trait entanglement.

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. The submission focuses on an interesting problem, whether we can use base models to make predictions about post-trained behaviour. The authors had some good ideas for experiments, although the fact that the authors choose a specific persona undermines conclusions. For the author’s conclusions to be supported, I would want to see the predictive power of base models beyond post-trained models.

  2. The central idea in this work is that identifying the persona responsible for some model output will help us to predict future behaviour. While this intuition seems sound to me, there are several aspects of the work that I struggle to understand. In particular: (i) What's the advantage of using base models, specifically, to infer personas from posttrained model outputs? (ii) What exactly is the sense in which persona inferences predict misalignment - in which contexts should we expect misaligned behaviour? (iii) What is the finding about stated and revealed preferences? The work could be improved by providing clearer answers to these questions.

Cite this project

@misc{tagade2026base,
  title = {{Base Model Persona Inference can Predict Misalignment and Surface Agreement}},
  author = {Arush Tagade and Taslim Mahbub},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/base-model-persona-inference-can-predict-misalignment-and-surface-agreement-ua8o}},
  url = {https://apartresearch.com/sprints/projects/base-model-persona-inference-can-predict-misalignment-and-surface-agreement-ua8o}
}

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