Self-assessment, Exhibition, and Recognition: a Review of Personality in Large Language Models

Reviews, theory, and governance2024arXivApproved editorial review

Authors: Zhiyuan Wen, Yu Yang, Jiannong Cao, Haoming Sun, Ruosong Yang, Shuaiqi Liu

Keywords: personality in LLMs, literature review, self-assessment, personality exhibition, personality recognition, research taxonomy, open resources

Source: Open primary source (opens in a new tab)

6
Authors
4
Findings
4
Limitations
3
Evidence

Editorial summary

English

This review organizes research on personality in LLMs into three problems: self-assessment, exhibition, and recognition. Self-assessment covers work that administers inventories to a model itself; exhibition separates methods that edit model parameters from approaches that induce behavior through prompting, fine-tuning, or related controls; recognition includes direct personality inference and LLM-enhanced recognition systems. The paper states that it reviews 72 studies published since 2022 through June 2024 and provides comparative tables of instruments, models, methods, code, and datasets, together with summaries of conflicting findings, open challenges, and possible applications. Its main value is taxonomic and navigational: it helps readers locate research lines, terminology, resources, and recurring tensions involving prompt sensitivity, validity of human instruments, stability, trait control, and ethics. It is not, however, reported as a reproducible systematic review. The paper does not provide databases, complete search strings and dates, inclusion and exclusion criteria, duplicate screening, extraction procedures, quality or risk-of-bias assessment, or evidence grading. Its claim to be the first comprehensive survey is not established through a verifiable comparison. The corpus is dominated by computer-science publications and preprints, and the authors themselves note insufficient grounding in social science. Resource links are not an archived, versioned artifact or reproduction protocol. The article should therefore be used to navigate the field as it stood in June 2024, not to estimate effects, prevalence, or cumulative certainty, and it is already a historical snapshot given the field’s subsequent rapid expansion.

Español

Esta revisión organiza la literatura sobre personalidad en LLM en tres problemas: autoevaluación, exhibición y reconocimiento. La autoevaluación agrupa trabajos que administran inventarios al propio modelo; la exhibición separa métodos para editar parámetros o inducir conductas mediante prompts, fine-tuning y otros controles; y el reconocimiento incluye inferencia directa de personalidad y sistemas mejorados con LLM. El texto declara revisar 72 trabajos desde 2022 hasta junio de 2024 y aporta tablas comparativas de instrumentos, modelos, métodos, código y conjuntos de datos, además de sintetizar resultados contradictorios, retos y posibles aplicaciones. Su valor principal es taxonómico y orientativo: permite localizar líneas de trabajo, vocabulario, recursos y tensiones como sensibilidad al prompt, validez de instrumentos humanos, estabilidad, control de rasgos y límites éticos. No obstante, no se presenta como una revisión sistemática reproducible en sentido metodológico. Faltan bases consultadas, cadenas de búsqueda, fechas y criterios completos, protocolo de inclusión y exclusión, cribado duplicado, proceso de extracción, evaluación de calidad o riesgo de sesgo y graduación de evidencia. La afirmación de ser la primera revisión exhaustiva no queda demostrada mediante una comparación verificable. La muestra está dominada por informática y preprints, y los propios autores reconocen una integración insuficiente de ciencias sociales. Los enlaces de recursos tampoco forman un archivo versionado ni un protocolo de reproducción. Por tanto, el artículo sirve para navegar el campo tal como era en junio de 2024, pero no para estimar efectos, prevalencias o certeza acumulada, y ya debe tratarse como una fotografía histórica frente al rápido crecimiento posterior.

Research question

How can the emerging literature on personality in LLMs be organized, and what are its main methods, findings, resources, challenges, and applications?

Method

Narrative review of 72 works up to June 2024, organized into self-assessment, exhibition, and recognition, with methodological subcategories and tables of instruments, models, code, and data.

Sample: Seventy-two publications reported from 2022 to June 2024; no selection diagram or reproducible screening protocol is reported.

Findings

  • The three-axis taxonomy integrates measurement, expression/control, and recognition.
  • The tables bring together instruments, models, and resources useful for guidance.
  • The literature presents contradictory results on stability and validity.
  • The field needs greater psychometric and social science integration.

Limitations

  • No reproducible systematic search is documented.
  • There is no evaluation of quality, risk of bias, or certainty.
  • Computer science works and preprints predominate.
  • Coverage ends in June 2024 and ages rapidly.

What the study does not establish

  • It does not quantify effect sizes or prevalences.
  • It does not allow concluding which method is superior with graded evidence.
  • It does not demonstrate exhaustiveness or historical priority.
  • It does not convert linked resources into archived and reproducible artifacts.

Traceability

Scope: Full text

Version: arXiv:2406.17624v1; complete 19-page PDF and TeX source; literature snapshot through June 2024

Consulted source: https://arxiv.org/abs/2406.17624v1

Review: Codex full-text, visual, taxonomy and review-method audit, 2026-07-18

Approval: Codex fidelity pass, 2026-07-18

English translation: approved, 2026-07-18

Models evaluated

  • Multiple LLM families covered narratively

Instruments and metrics

  • Big Five inventories
  • MBTI-related instruments
  • Dark-trait, value and other personality measures

Data used

  • Public resources listed across 72 reviewed papers

Evidence and location

  • Taxonomy and declared scope: arXiv v1, sections 1–2 and Figure 1
  • Self-assessment, exhibition, and recognition: arXiv v1, sections 3–5 and comparison tables
  • Challenges, resources, and future directions: arXiv v1, sections 6–8 and appendices