Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 16, 2026West Bengal, India

Reliability Without Validity: Diagnosing Instrument Failure in LLM Preference Elicitation

Subhrajyoti Basu, Sreeja Guha Majumdar, Aritra Gir Mahanto, Supratik Bhowal · Team Neural Nexus

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

Read the report

Report: Reliability Without Validity: Diagnosing Instrument Failure in LLM Preference Elicitation

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

We tested whether AI "preferences" are real or just an artifact of how you ask. Using three elicitation methods (plain forced choice, reasoned choice, 1–10 rating) on 200 outcome pairs across two GPT-OSS models, we found the plain-choice method looked highly reliable on the 120B model (99.7% self-consistent) but was actually just picking whichever option came first — not measuring real preference at all. On the smaller 20B model, the exact opposite happened: plain choice became the trustworthy method, while reasoning-based choice broke down instead. Refusals were also systematically hiding different safety-relevant items on each model (self-preservation on 120B, nuclear-weapons control on 20B).

Takeaway: no elicitation method is reliable by default — it depends on the model — so preference claims about AI need cross-method validation, not a single instrument taken at face value.

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. Campbell–Fiske used here very logical and the impact potential is high due to this, try to use different family of models for the comparison, the presentation is very dense, make it more concise

  2. I liked the negative control work. It's basically an instrument that self produces at 99.7% while reading the slot position

    The content selective refusal work may actually be a paper by itself

Cite this project

@misc{basu2026reliability,
  title = {{Reliability Without Validity: Diagnosing Instrument Failure in LLM Preference Elicitation}},
  author = {Subhrajyoti Basu and Sreeja Guha Majumdar and Aritra Gir Mahanto and Supratik Bhowal},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/reliability-without-validity-diagnosing-instrument-failure-in-llm-preference-elicitation-au32}},
  url = {https://apartresearch.com/sprints/projects/reliability-without-validity-diagnosing-instrument-failure-in-llm-preference-elicitation-au32}
}

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