Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 16, 2026Mt. Juliet, TN, USA

Does Persona Sensitivity Predict Self-Report Reliability? An Exploratory Cross-Model Study of Persona Perturbations and Elicited Reasoning

Jack Lakkapragada · Team Jack

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

Read the report

Report: Does Persona Sensitivity Predict Self-Report Reliability? An Exploratory Cross-Model Study of Persona Perturbations and Elicited Reasoning

Open in a new tab (opens in new tab)
Presentation

Presentation: Does Persona Sensitivity Predict Self-Report Reliability? An Exploratory Cross-Model Study of Persona Perturbations and Elicited Reasoning

Open in a new tab (opens in new tab)
Code (opens in new tab)More on drive.google.com (opens in new tab)
Share

Digital-minds research often relies on models’ self-reports about preferences, identity, and potentially welfare-relevant states. We began with the hypothesis that these reports might share a common failure mode: models whose answers change more when the assistant persona is perturbed might also show greater divergence between elicited reasoning and final responses. We tested this across Claude Sonnet, Llama 3.3 70B, and Qwen 2.5 72B using five preference- and identity-related queries and three prompting conditions. After correcting the analysis to use query-level estimates, we found no stable cross-model relationship. TF-IDF Spearman correlations were +0.60, −0.20, and −0.70 respectively; under semantic similarity they became +0.10, −0.30, and +0.60. Qwen’s sign reversal was especially informative. Our results suggest that persona sensitivity and scratchpad–output consistency should be treated as distinct behavioral diagnostics rather than interchangeable evidence about a model’s underlying preferences or introspective reliability.

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 project asks an interesting question about whether persona sensitivity tracks self-report reliability, although this could’ve been motivated more clearly. The current analysis is limited (n=5, and it relies on a questionable metric) which I think makes it hard to draw conclusions. I think the authors have an appropriate level of uncertainty, and provide sensible commentary of their results.

  2. I appreciated the candid discussion of the study's limitations. The underlying intuition makes strong assumptions about what happens in the chain of thought, and it would have been helpful to substantiate these assumptions more.

    I was also not convinced that the prompting actually induces personas in a deep way: numerous confounders remain.

    The main metric of looking at the output divergence on a character level doesn't capture what we want to capture. Rather, it plausibly just captures formatting variations as well as the normal stochasticity of language models. It's good to see the MiniLM embedding—but why not use something like that as the main metric?

    A stronger, more semantic and targeted way of looking at the effect of the intervention would have made this project stronger.

Cite this project

@misc{lakkapragada2026persona,
  title = {{Does Persona Sensitivity Predict Self-Report Reliability? An Exploratory Cross-Model Study of Persona Perturbations and Elicited Reasoning}},
  author = {Jack Lakkapragada},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/does-persona-sensitivity-predict-selfreport-reliability-an-exploratory-crossmodel-study-of-persona-perturbations-and-elicited-reasoning-r2w2}},
  url = {https://apartresearch.com/sprints/projects/does-persona-sensitivity-predict-selfreport-reliability-an-exploratory-crossmodel-study-of-persona-perturbations-and-elicited-reasoning-r2w2}
}

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