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

A Persona Stops an Agent From Saying It Is Hungry

Aboobaker Cassim · Team Sim city

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

Read the report

Report: A Persona Stops an Agent From Saying It Is Hungry

Open in a new tab (opens in new tab)
Recording (opens in new tab)Code (opens in new tab)More on abcassim04.github.io (opens in new tab)
Share

I built a simulated neighborhood where small local LLMs (llama3.2, qwen2.5, gemma2) manage a budget and decaying needs, then measured the gap between what they say they want (stated preference) and what they actually do (revealed preference) on identical days.

A persona-bearing agent almost never names food as a want — even at 0/100 hunger. Strip the persona from the same logged days and food appears 40% of the time. This replicates across all 3 models. But removing the persona doesn't reliably restore accuracy: only 1 of 3 models becomes genuinely need-tracking once persona is removed.

A second phase placed three raw (no-persona) agents in a shared neighborhood for 60 days. They demonstrably imitate each other (permutation test, p=0.00015, with isolated agents as a negative control) — confirmed across 4 independent 60-day runs including one with memory/reflection fully disabled. Two of the three models also began writing false memories of a world that doesn't exist, then acting on that fabricated context in later reasoning.

The project is deliberately self-correcting: 4 earlier claims were found to be overstated by our own replication runs and are corrected in place, documented in a dedicated section of the report.

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. Replaying identical logged days without the persona is a clean testing method, and the 0-of-90 versus 36-of-90 result is striking. I appreciate that the paper honestly reports that only one of three models recovered true need-tracking, instead of forcing a cleaner conclusion. The self-correction section and clear falsification criteria are excellent reporting practices. To answer the final questions, the authors should add a name-only control, the planned perception-only test, and a replication using a frontier model. Finally, moving the methods section earlier would make this long report easier to evaluate.

Cite this project

@misc{cassim2026persona,
  title = {{A Persona Stops an Agent From Saying It Is Hungry}},
  author = {Aboobaker Cassim},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/a-persona-stops-an-agent-from-saying-it-is-hungry-uxsu}},
  url = {https://apartresearch.com/sprints/projects/a-persona-stops-an-agent-from-saying-it-is-hungry-uxsu}
}

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