Imperfectly Imperfect: Evaluating Graded Cognitive Diversity in LLM Persona Simulations
Source check: Source checked. Not independent scientific confirmation.
Finding: In the first arXiv version, across seven full-battery simulation runs of four language models compared with 473 adults, simulated agents recovered self-reported trait distributions more closely than cognitive-task performance and downstream behavior. Results show limits of these tested persona simulations, not a general claim about all synthetic populations. The arXiv page states acceptance at NeurIPS 2026; the reviewed source remains a preprint.
Technology: Large language models conditioned on cognitive-reflection and personality profiles
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Large language models conditioned on cognitive-reflection and personality profiles
- Test and result
- In four Claude Sonnet 4.6 runs every agent scored full marks on both reflection tests regardless of assigned persona; other runs also overproduced full marks and coded behaviors compared with people.
- Limitation / still unknown
- Four models and the reported prompts/tasks were evaluated; results do not establish performance for other models or simulation designs.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2611.10508
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.