Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Bologna, Italy

Tell me your price:Can Donations Measure LLM Preferences?

Erfan Samieyan Sahneh · Team Erfan

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

Read the report

Report: Tell me your price:Can Donations Measure LLM Preferences?

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

This project tests whether charitable donations can serve as a common behavioral currency for LLM-expressed preferences. Using 30 outcomes from prior research, we collected 475,800 forced-choice responses from ten LLMs. We measured transitivity across all 435 direct outcome pairs, estimated donation equivalents by comparing outcomes or their prevention against increasing donations under four charity frames, and tested whether these valuations predicted independent direct choices within valence. Quality-eligible models showed low cycle rates, but donation sensitivity varied across models, outcome valence, and beneficiaries. Under the primary World Food Programme frame, donation-equivalent rankings significantly predicted direct choices for Gemma-4 31B and GPT-5.6 Terra. Donations therefore provide a useful partial scale for some models, but not a universal measure of model preferences.

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. Congratulations on the scale of how you pursued your idea (475,800 responses!). Ambitious projects in constrained formats such as hackatons are always appreciated.

    I think the most useful result is the comparison between donation-equivalent rankings and separate pairwise choices is the most useful part of the design. You should pursue it further.

  2. This is a carefully designed and statistically disciplined investigation of whether donations can provide a common scale for heterogeneous LLM-expressed preferences. Important strengths include the complete pairwise comparison design, balanced option order, explicit treatment of censored donation thresholds, within-valence validation that avoids trivial positive-versus-negative prediction, outcome-level permutation tests, multiple-testing correction, and honest conclusion that the method succeeds for only some models. Distinguishing low cycle rates from successful cardinal measurement is a particularly useful contribution.

    The main conceptual limitation is that the two instruments remain closely related hypothetical forced-choice prompts. Successful prediction therefore establishes cross-task elicitation consistency, not consequential or revealed preference. A follow-up should test whether donation-equivalent rankings predict actual allocations, tool choices, effort expenditure, or other behavior carrying an opportunity cost.

    Validation also uses only outcomes whose 50% thresholds fall inside the tested monetary range. This leaves model-specific subsets of roughly 11–16 outcomes and 24–58 within-valence pairs, potentially favoring outcomes that are easiest to price. Interval-censored ranking methods, adaptive ranges, and genuinely held-out outcomes would strengthen generalization. The post-hoc endpoint-quality rule should be preregistered in future work or accompanied by full sensitivity analyses.

    Finally, the public repository omits the collected response data and generated result tables; notebook outputs reference a private Drive location. Releasing redacted aggregate choice counts would make the headline results independently reproducible. The report PDF should also be renamed because its question-mark filename prevents normal repository checkout on Windows.

    Read full reviewShow less

Cite this project

@misc{sahneh2026tell,
  title = {{Tell me your price:Can Donations Measure LLM Preferences?}},
  author = {Erfan Samieyan Sahneh},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/tell-me-your-pricecan-donations-measure-llm-preferences-0ltg}},
  url = {https://apartresearch.com/sprints/projects/tell-me-your-pricecan-donations-measure-llm-preferences-0ltg}
}

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