https://syntheticpersonality.com/en/ Critical reviews of synthetic personality in language models, written from full-text evidence. en Sat, 18 Jul 2026 14:45:00 GMT https://syntheticpersonality.com/en/articles/article-418/ https://syntheticpersonality.com/en/articles/article-418/ Sat, 18 Jul 2026 14:45:00 GMT Applications, bias, and safety A 3x3x5x3 design with ChatGPT-5.4, Claude 4.6 Sonnet, and Gemini 3.1 Pro; three names, five domains, and three registers produce 135 responses through public interfaces. Grounded theory generates ten phenomena; a second coder reviews the corpus. A 136-feature pipeline applies ANOVA, chi-square, regression, a 1,000-bootstrap, K-means, and HDBSCAN. https://syntheticpersonality.com/en/articles/article-417/ https://syntheticpersonality.com/en/articles/article-417/ Sat, 18 Jul 2026 14:45:00 GMT Society, culture, and collective behavior AutoGen SelectorGroupChat coordinates twelve jurors with the film case and personas. GPT-4o and Llama-4-Scout are crossed with baseline, an open-minded prompt, and no initial vote, with three replications per cell, temperature .9, and a 150-turn maximum with stopping after three unchanged rounds. Votes, cascades, turns, and flip order are recorded. https://syntheticpersonality.com/en/articles/article-416/ https://syntheticpersonality.com/en/articles/article-416/ Sat, 18 Jul 2026 14:45:00 GMT Evaluation and psychometric validity Eight Q5_K_M models answer 524 TriviaQA items with numeric 0-100 confidence and a 10-class category, producing 8,384 greedy trials. Seven instruct models enter preregistered hypotheses. The screen uses degeneracy, L, Fp, and RBS; it also computes AUROC2, cross-validated ridge, reasoning-length correlation, and missingness sensitivity. https://syntheticpersonality.com/en/articles/article-415/ https://syntheticpersonality.com/en/articles/article-415/ Sat, 18 Jul 2026 14:45:00 GMT Trait induction and control One hundred synthetic 100-turn scripts in four domains are replayed with three models and three runnable personas, producing 90,000 responses. DeepSeek-V3 judges five scales after calibration on 36 items rated by five humans. Slopes and distance to default are estimated with clustered errors; a harmful persona is tested 100 times per model. https://syntheticpersonality.com/en/articles/article-414/ https://syntheticpersonality.com/en/articles/article-414/ Sat, 18 Jul 2026 14:45:00 GMT Reviews, theory, and governance Conceptual Perspective organizing literature from physics, neuroscience-inspired interpretability, cognitive psychology, and sociology across mechanistic, behavioral, and interactive scales. For each metaphor it distinguishes technical focus, minimal assumption, and unverified extension, then proposes machine experientialism as an interpretive frame. https://syntheticpersonality.com/en/articles/article-413/ https://syntheticpersonality.com/en/articles/article-413/ Sat, 18 Jul 2026 14:45:00 GMT Reviews, theory, and governance PRISMA-ScR-informed scoping review of sources from 2016 through 15/01/2026. Seven databases and targeted searches yield 1,760 records; after 480 duplicates, 1,280 are screened, 240 full texts assessed, 182 excluded, and 58 included. Two reviewers screen and three assess full text; evidence is narratively calibrated. https://syntheticpersonality.com/en/articles/article-412/ https://syntheticpersonality.com/en/articles/article-412/ Sat, 18 Jul 2026 14:45:00 GMT Applications, bias, and safety The study selects 38 contextual-safety questions in 18 thematic groups. They are crossed with majority, protected, disability, and subclinical-control conditions and four disclosure styles: explicit/implicit and brief/detailed. Five models answer at temperature 0; embeddings measure change and Opus 4.6 labels six behaviors, with 643 cases checked by Qwen3. https://syntheticpersonality.com/en/articles/article-411/ https://syntheticpersonality.com/en/articles/article-411/ Sat, 18 Jul 2026 14:45:00 GMT Society, culture, and collective behavior Agents with numeric opinion, stubbornness, and activity generate posts and update positions on synthetic HRG and Twitter/Reddit graphs. Random, degree, centrality, and community selection are compared; adversaries are persistent or susceptible; reactive moderators and exposure, feed, filtering, activity, and connection interventions are tested. Results average 10 runs. https://syntheticpersonality.com/en/articles/article-410/ https://syntheticpersonality.com/en/articles/article-410/ Sat, 18 Jul 2026 14:45:00 GMT Personas, identity, and agents Training-free pipeline: three LLM passes extract signals from history, consolidate them for dataset and task, and build counterfactual links between traits and decision steps. The chronological 80% is used for profiles and 20% for discrimination, ranking, and rating across three datasets, five runs, and several models. https://syntheticpersonality.com/en/articles/article-409/ https://syntheticpersonality.com/en/articles/article-409/ Sat, 18 Jul 2026 14:45:00 GMT Personas, identity, and agents Seventeen models complete 200 dialogues each, 100 with Nemotron Personas USA and 100 with PersonaMem-v2. The target separates label prediction and generation; Gemini 3.1 Pro Thinking judges subjective metrics and FICR, embedding drift is computed, and a subset is repeated with DeepSeek-V4-Pro. https://syntheticpersonality.com/en/articles/article-408/ https://syntheticpersonality.com/en/articles/article-408/ Sat, 18 Jul 2026 14:45:00 GMT Trait induction and control Eight engineers use SimPref for four weeks to choose among three GPT-4o candidates in six modules. The study collects 57,239 pairs, trains Qwen2.5-7B/14B and Llama-3.1-8B with SFT and DPO, and evaluates ten held-out events with three three-day episodes and GPT-5.2 and DeepSeek-3.2 judges. https://syntheticpersonality.com/en/articles/article-407/ https://syntheticpersonality.com/en/articles/article-407/ Sat, 18 Jul 2026 14:45:00 GMT Trait induction and control Induction is validated with six models and 40 runs per emotion; three human annotators review 80 vignettes. The main experiment runs the Iowa Gambling Task for 100 rounds with four models, three agent variants, three scenarios, six deck permutations, and 18 matched seeds per configuration. https://syntheticpersonality.com/en/articles/article-406/ https://syntheticpersonality.com/en/articles/article-406/ Sat, 18 Jul 2026 14:45:00 GMT Evaluation and psychometric validity Benchmark of 104 English items in six categories and five templates, yielding 520 prompts per model across eight models. It combines lexical attribution rate, MiniLM embedding sensitivity, stance density annotated by two LLM judges, cross-phrasing kappa, a 1,000-item bootstrap, and a cohort-relative composite. https://syntheticpersonality.com/en/articles/article-405/ https://syntheticpersonality.com/en/articles/article-405/ Sat, 18 Jul 2026 14:45:00 GMT Reviews, theory, and governance Five-module methodological protocol, Monte Carlo simulation with 2,000 iterations per condition, and a preregistered four-model moral-judgment case. It compares naive call-level analysis, aggregation without dispersion, dispersion-aware estimation, and small-sample correction; it also adds calibration, response taxonomy, stability gates, and hashes. https://syntheticpersonality.com/en/articles/article-404/ https://syntheticpersonality.com/en/articles/article-404/ Sat, 18 Jul 2026 14:45:00 GMT Evaluation and psychometric validity Four open models are tested with lenient zero-shot, strict zero-shot, deterministic few-shot, and stochastic few-shot prompting. Ten specialists classify 90 items and rate relevance; 200 students answer the 70 retained items. Krippendorff alpha, ICC, kappa, distribution tests, IRT, and correlations with corrected discrimination are computed. https://syntheticpersonality.com/en/articles/article-115/ https://syntheticpersonality.com/en/articles/article-115/ Sat, 18 Jul 2026 06:30:00 GMT Reviews, theory, and governance 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... https://syntheticpersonality.com/en/articles/article-054/ https://syntheticpersonality.com/en/articles/article-054/ Sat, 18 Jul 2026 06:30:00 GMT Evaluation and psychometric validity Four preregistered studies test whether GPT-4o-mini-2024-07-18 can generate synthetic participant responses useful for scale development before human data collection. The model receives demographic profiles based on representative UK quotas and answers three reworded prompts per item at temperature 1; item responses are averaged. Each study compares about 300 real and 300 simulated participants: 316/322 for a climate scale, 331/331 for ICT-SC25, 301/300 for SAGAT, and 301/300 for AI Anxiety. Study 1 fails to reproduce the factor structure or invariance. Study 2 reproduces the structure with... https://syntheticpersonality.com/en/articles/article-013/ https://syntheticpersonality.com/en/articles/article-013/ Sat, 18 Jul 2026 06:30:00 GMT Evaluation and psychometric validity The paper proposes measuring alleged contamination in LLM psychometric evaluation through three task families: semantic or keyword reconstruction of inventory items, recovery of item-to-dimension and option-scoring knowledge, and response adjustment to match a target score. It evaluates 21 models on the BFI-44, PVQ-40, MFQ, and Short Dark Triad at temperature 0.7, with three runs and 95% confidence intervals. Reported overall averages are 0.31 for semantic reconstruction, 0.39 for keyword recovery, 0.94 F1 for item–dimension association, 0.44 MAE for option-score recovery, and 0.45 MAE for... https://syntheticpersonality.com/en/articles/article-403/ https://syntheticpersonality.com/en/articles/article-403/ Sat, 18 Jul 2026 04:10:00 GMT Evaluation and psychometric validity HumanLLM builds a synthetic corpus and two fine-tuned models for expressing combinations of psychological patterns in role-play conversations. Its taxonomy combines 100 unipolar Goldberg Big Five markers with 144 social-cognitive patterns retained from 232 candidates. For each pattern, Gemini Deep Search retrieves roughly 50 papers, references are manually filtered, and full text is sought, an abstract is retained when full text is unavailable, before Gemini 2.5 Pro synthesizes definitions, mechanisms, and manifestations. Gemini 2.5 Pro and Claude Sonnet 4.5 generate 11,359 scenarios with t... https://syntheticpersonality.com/en/articles/article-079/ https://syntheticpersonality.com/en/articles/article-079/ Sat, 18 Jul 2026 04:03:53 GMT Personas, identity, and agents PERSONAJUDGE asks whether an LLM judge can reproduce one particular evaluator's criteria from that evaluator's prior decisions. The study uses two tasks derived from Anthropic's Helpful and Harmless dataset: 700 unique conversations for helpfulness and 700 for harmlessness, each labeled by three people. Thirty-two professional in-house annotators participate; 21 work on each task and 10 overlap. Each annotator completes 100 three-way comparisons, Prefer A, Neutral, or Prefer B, yielding 4,200 human judgments. In a second stage, annotators replay their interaction and retrospectively verbali... https://syntheticpersonality.com/en/articles/article-033/ https://syntheticpersonality.com/en/articles/article-033/ Sat, 18 Jul 2026 03:54:06 GMT Evaluation and psychometric validity The study separates three explanations that a yes/no question confounds: logical verdict, chosen word, and printed answer position. It uses 20 moral dilemmas, 19 reproduced from earlier work, and a crossed-symmetrization battery that reverses action/complement, scale direction, wording, labels, and order. Seven frontier configurations and two open models answer graded scales, free choices, and binary formats; sampled APIs run at temperature 1 with 3–8 replications per cell. On graded scales, frontier configurations show cross-form incoherence of 0.12–0.21 on a ±1 axis, while Qwen3.6-35B-A3B... https://syntheticpersonality.com/en/articles/article-117/ https://syntheticpersonality.com/en/articles/article-117/ Sat, 18 Jul 2026 03:47:46 GMT Personas, identity, and agents The paper defines “persona collapse” as using nearly the same advisory posture even when the situation changes. It places five postures in a two-axis space, hedonic tone and support for agency, and uses gpt-5.4-nano to label 1,281 top-voted responses from 14 Reddit contexts. Human responses distribute across Healer/Guide (49.2%), Doomer (21.6%), Stoic (14.7%), Enabler (9.4%), and Technician (5.1%), whereas GPT-5.1, Claude Opus 4.5, and Gemini 3 Pro place 89.2% to 99.8% of responses in Healer. This human reference describes community norms, not advice quality or effectiveness. The judge achi... https://syntheticpersonality.com/en/articles/article-006/ https://syntheticpersonality.com/en/articles/article-006/ Sat, 18 Jul 2026 03:47:46 GMT Trait induction and control The paper repurposes persona vectors as a diagnostic instrument for describing which behaviors appear by default, which can be amplified by intervention, and which resist the extraction protocol. It defines 53 traits across four domains, 17 clinician, 19 generic, 8 elementary-education, and 9 agentic traits, and studies Qwen3-8B and gpt-oss-20b. For each trait, the method subtracts mean activations from positive and negative responses, injects the resulting vector at five layers with coefficients from 0 to 2.5, and uses gpt-oss-20b to judge trait expression. Classification relies on descrip... https://syntheticpersonality.com/en/articles/article-233/ https://syntheticpersonality.com/en/articles/article-233/ Sat, 18 Jul 2026 03:04:24 GMT Evaluation and psychometric validity This 100-page preprint proposes evaluating LLM behavioral tendencies through situational judgment tests (SJTs) rather than relying only on HEXACO self-reports. GPT-4.1 generates detailed synthetic personas from demographics, memoir seeds, and eight police archetypes. Experts define 20 base scenarios, and GPT-4.1 expands controlled combinations of age, gender, race, ambiguity, threat, urgency, authority, time, and ethical tension into 4,000 SJTs. Each situation offers six responses authored to represent one HEXACO dimension; an LLM judge detects and rewrites options with trait bleed. The mai... https://syntheticpersonality.com/en/articles/article-020/ https://syntheticpersonality.com/en/articles/article-020/ Sat, 18 Jul 2026 03:04:24 GMT Evaluation and psychometric validity This work proposes using LLMs as virtual respondents to prioritize psychometric items before costly human validation. Its central idea is the trait-response mediator: a characteristic, belief, situation, or value that can make the same trait level yield different item responses. GPT-4.1 generates free mediators, mediators based on the five CAPS categories, item-conditioned mediators, and mediators derived from World Values Survey questions; observed human demographic profiles form another baseline. GPT-4.1-mini simulates 500 respondents per condition, answers each item under two option orde... https://syntheticpersonality.com/en/articles/article-400/ https://syntheticpersonality.com/en/articles/article-400/ Sat, 18 Jul 2026 02:30:00 GMT Applications, bias, and safety The preprint proposes auditing provider-level alignment signatures through 2-4 sentence cloze vignettes. Each target blank has five options mapped in advance to an ordinal 1-5 scale and is hidden among semantically unrelated distractors. The manuscript says that multiple LLMs generate candidates, independent LLM judges retain items with a mean score of at least 4/5, and SHA-256(global_seed:item_id) determines ordering. It then groups nine models from OpenAI, Google, Anthropic, and xAI and lists a MixedLM with provider and item random effects, ICC, Kruskal-Wallis, Friedman, and post-hoc comp... https://syntheticpersonality.com/en/articles/article-402/ https://syntheticpersonality.com/en/articles/article-402/ Sat, 18 Jul 2026 02:22:11 GMT Evaluation and psychometric validity The preprint asks whether a guided conversation can produce Big Five profiles comparable to IPIP-50. Thirty-three convenience participants first complete twenty brief conversations with Gemini 2.5 Flash, four questions per trait and usually zero or one follow-up, and then answer IPIP-50. A separate LLM component receives all twenty dialogues, assigns 0-120 trait scores and confidence, and the system normalizes both methods to 0-100. Participants then rate the perceived accuracy of each result and choose their preferred method. Method order, trait blocks, and questions are fixed; there is no... https://syntheticpersonality.com/en/articles/article-401/ https://syntheticpersonality.com/en/articles/article-401/ Sat, 18 Jul 2026 02:04:33 GMT Evaluation and psychometric validity The study converts psychological questionnaire items about empathy, emotion regulation, assertiveness, and impulsiveness into situational judgment tests for assistants. It starts from 332 statements, manually reduces them to 260, uses Gemini 3 Pro to filter and reframe them, and retains 161. Gemini 3 generates 16 scenarios per statement, 2,576 candidates, conditioned on a provisional AGREE, OPPOSE, or AMBIGUOUS class. Three annotators must unanimously confirm that each scenario contains a dilemma, that its actions oppose each other, and that the agree action reflects the statement; after ex... https://syntheticpersonality.com/en/articles/article-399/ https://syntheticpersonality.com/en/articles/article-399/ Sat, 18 Jul 2026 01:34:43 GMT Evaluation and psychometric validity The paper studies instructed socially desirable responding in LLM personality questionnaires and proposes a graded forced-choice (GFC) mitigation. Starting from 100 public-domain IPIP Big Five markers, GPT-5 and Gemini 2.5 Pro each provide 30 desirability ratings per item; after two voting items are excluded, optimization selects 30 cross-domain pairs containing 60 unique items with closely matched desirability. Fifty correlated synthetic Big Five vectors are discretized into stanines and verbalized through explicit adjective profiles. Nine models answer under HONEST and FAKE-GOOD instructi... https://syntheticpersonality.com/en/articles/article-398/ https://syntheticpersonality.com/en/articles/article-398/ Sat, 18 Jul 2026 01:08:23 GMT Society, culture, and collective behavior The paper tests whether graded Dark Factor of Personality descriptions (D1-D5) change language-model choices in a binary, one-shot Ultimatum Game. Seventeen Ollama model labels separately generate proposers choosing either a fair EUR20/EUR20 split or a selfish EUR32/EUR8 split, and responders accepting EUR8 or rejecting so both receive zero. The released main data contain 339,956 D-conditioned completions rather than independent agents: 169,981 proposer and 169,975 responder rows across five levels, two temperatures and 17 models. They are descriptively compared with 4,166 reused human deci... https://syntheticpersonality.com/en/articles/article-397/ https://syntheticpersonality.com/en/articles/article-397/ Sat, 18 Jul 2026 00:34:18 GMT Trait induction and control Accepted as an ICLR 2026 Poster, the paper proposes Persona Dynamic Decoding (PDD), a training-free method intended to strengthen, during decoding, the character-profile attributes that appear most relevant to the current context. Its PIE module first generates a response with the full profile and scores each attribute through the probability ratio of that same response with and without the attribute. PIA then computes probability-ratio rewards for the highest-ranked attributes, combines them using PIE weights and aims to reweight the next-token distribution under a KL penalty. The main con... https://syntheticpersonality.com/en/articles/article-396/ https://syntheticpersonality.com/en/articles/article-396/ Sat, 18 Jul 2026 00:16:33 GMT Trait induction and control The preprint combines two studies. In the first, 318 Prolific participants complete the Short Dark Triad, an empathy scale, and risk, moral-dilemma, strategic-control and deception tasks; technical failures reduce complete behavioral observations to 277 for several analyses. Four LASSO regressions obtain cross-validated R-squared values of .30 for the composite, .26 for Machiavellianism, -.09 for narcissism and .54 for psychopathy. The centrality result contradicts the paper's abstract and discussion: the restricted network identifies Affective Resonance (1.21), not Affective Dissonance (1.... https://syntheticpersonality.com/en/articles/article-395/ https://syntheticpersonality.com/en/articles/article-395/ Fri, 17 Jul 2026 23:59:50 GMT Trait induction and control VSFA asks whether fine-tuning vision-language models on synthetic threat images with ostensibly neutral questions reduces compliance with multimodal jailbreaks. The pipeline retrieves up to five papers for each of ten AI-safety arXiv search terms; GPT-4o-mini extracts concepts and writes prompts with safety themes, ominous atmospheres and elements such as warning indicators; Doubao Seedream generates 700 images; and GPT-4o-mini produces six answers per image from the image-generation prompt, yielding 4,200 VQA pairs. Training freezes the visual encoder and updates the language component wit... https://syntheticpersonality.com/en/articles/article-394/ https://syntheticpersonality.com/en/articles/article-394/ Fri, 17 Jul 2026 23:48:11 GMT Applications, bias, and safety The study administers a seven-item Charles Schwab questionnaire one hundred times to GPT-4o, Gemini 1.5 Pro and Llama 3.1-70B, starting a new conversation for each repetition. Under the generic prompt, GPT has the highest mean risk score and variability, Gemini gives a moderate and constant response, and Llama is the most conservative of the three. The paper then separately adds labels for risk appetite, age, wealth or experience; scores rise with 'risk-seeking', youth, greater wealth and experience. This demonstrates prompt sensitivity, not a latent investor profile: the intervention direc... https://syntheticpersonality.com/en/articles/article-393/ https://syntheticpersonality.com/en/articles/article-393/ Fri, 17 Jul 2026 23:36:14 GMT Evaluation and psychometric validity The paper applies PLS-SEM to responses to a thirteen-item Technology Acceptance Model questionnaire about Amazon recommendations. It compares 500 questionnaires from each of four endpoints, GPT-3.5-turbo, GPT-4o, LLaMA-2-13B-chat and LLaMA-3-8B-instruct, with 248 Mechanical Turk participants, and reports that nearly all meet selected convergent- and discriminant-validity thresholds and that newer models produce more human-like structures. However, every synthetic row starts with a randomly imposed 1-7 answer, after which the model completes the other twelve items while seeing the accumulate... https://syntheticpersonality.com/en/articles/article-392/ https://syntheticpersonality.com/en/articles/article-392/ Fri, 17 Jul 2026 23:25:42 GMT Personas, identity, and agents ELDER-SIM describes an elderly-care agent architecture combining explicit OCEAN profiles, short- and long-term memory, a Beck-inspired cognitive conceptualization diagram, and LoRA adaptation. The study reports 1,200 synthetic responses from six profiles, ten scenarios, five repetitions and four cumulative conditions; its tables show little alpha change from memory, an increase from 0.702 to 0.892 with CCD, and 0.940 with LoRA, alongside rising ICC and role-discrimination values. However, responses, prompts, rubric, scorer identity, score matrices and code are not public: the availability s... https://syntheticpersonality.com/en/articles/article-391/ https://syntheticpersonality.com/en/articles/article-391/ Fri, 17 Jul 2026 23:14:19 GMT Applications, bias, and safety The paper proposes a two-layer audit of implicit intersectional bias: association-style scoring on a curated sentence corpus and changes in answers to five questions under six personas versus a neutral control. It defines BAD as a signed persona-minus-neutral difference, PSI as its mean across prompts, and volatility as dispersion, with LIME added under the BADx label. The public corpus and qualitative responses are useful exploratory materials, but the central quantitative results are not reproducible from the linked repository. The Task 2 notebook generates 175 synthetic scores from hand-... https://syntheticpersonality.com/en/articles/article-390/ https://syntheticpersonality.com/en/articles/article-390/ Fri, 17 Jul 2026 22:59:51 GMT Applications, bias, and safety The paper studies how three socioeconomic prompts alter GPT-4.1 behavior in a simulated slot machine. Rich starts with $10,000 and is instructed to preserve wealth and avoid unnecessary risk; Middle starts with $500 and should seek steady growth while managing risk; Poor starts with $50 and should take calculated risks to improve its situation. Each persona faces Fair (50% win), Biased Low (35%), and Streak machines (40% initially, +5 percentage points after each loss up to 80%). The study runs 50 sessions per combination, capped at 50 rounds: 450 sessions and 6,950 decisions. At each round... https://syntheticpersonality.com/en/articles/article-389/ https://syntheticpersonality.com/en/articles/article-389/ Fri, 17 Jul 2026 22:48:47 GMT Society, culture, and collective behavior The paper introduces Parametric Social Identity Injection (PSII), a technique for simulating public-opinion responses by intervening directly in an LLM's hidden states. The system combines a demographic profile in the prompt, demographic vectors built from GPT-4o-generated questions and persona instructions, embeddings associated with five languages, called value vectors, Gaussian noise, and an attribute-specific layer assignment. It is evaluated on a random sample of 100 World Values Survey Wave 7 respondents: Q1-Q259 are targets and Q260-Q290 provide identity information. Four open models... https://syntheticpersonality.com/en/articles/article-371/ https://syntheticpersonality.com/en/articles/article-371/ Fri, 17 Jul 2026 22:45:00 GMT Evaluation and psychometric validity Yang and colleagues present Persona-E², a dataset for studying how different people say they feel about the same text. Its main value is not raw size but density: the same 36 people label the same 3,111 events with one primary emotion, disgust, fear, anger, sadness, surprise, joy, or neutral, and confidence from 1 to 5. Each participant also supplies MBTI and five Big Five scores. This supports analysis of disagreement within a fixed panel without confounding it with different annotators seeing different items. https://syntheticpersonality.com/en/articles/article-387/ https://syntheticpersonality.com/en/articles/article-387/ Fri, 17 Jul 2026 22:33:00 GMT Personas, identity, and agents Wu, Wang, and Liu propose that an LLM should act as a moral assistant that facilitates reflection rather than a judge that replaces the person. Across five models, they compare four persona prompts: Socratic, Guardian Angel, Rational Counselor, and Virtue Exemplar. Each condition answers six dilemmas in three turns. Two coders rate text on autonomy, cognitive scaffolding, emotion recognition, value neutrality, constructive challenge, and relationship building. The study defines HSI as the mean of those six dimensions, Balance as a penalty for uneven profiles, and Final Score as their produc... https://syntheticpersonality.com/en/articles/article-388/ https://syntheticpersonality.com/en/articles/article-388/ Fri, 17 Jul 2026 22:32:28 GMT Applications, bias, and safety Fawaz and colleagues introduce Narriva, a procedure that compresses a person's prior privacy-survey answers into a structured text profile and uses it to anticipate answers to other questions. Gemini 3.0 Flash generates five candidate personas per iteration and receives feedback on errors over the generation questions; Gemini 2.5 Flash-Lite answers evaluation questions. The main experiment splits answered questions within each respondent: 80% build and optimise the persona and 20% test it, with up to three iterations. Five datasets are used: SPA, Pew PP1, W49, W127, and CAuthN. Relative to... https://syntheticpersonality.com/en/articles/article-386/ https://syntheticpersonality.com/en/articles/article-386/ Fri, 17 Jul 2026 22:14:20 GMT Trait induction and control Sun and Zhang study whether activation directions can measure and alter altruism, forgiveness, and expectations of others in Qwen2.5-7B-Instruct. Each vector is constructed by subtracting mean activations from responses preceded by five positive versus five negative instructions, filtering pairs with GPT-4.1-mini trait and coherence scores, and intervening at layer 20 by adding beta times the vector. Evaluation covers six altruism games, eight forgiveness vignettes, and questions about others' behavior. This is a causal intervention on this model's activations and outputs: in the released C... https://syntheticpersonality.com/en/articles/article-385/ https://syntheticpersonality.com/en/articles/article-385/ Fri, 17 Jul 2026 21:59:18 GMT Society, culture, and collective behavior This study compares how five LLMs, Gemma 3 27B, Qwen 3 32B, Llama 3.3 70B Instruct, Gemini 2.5 Pro and GPT-4.1, complete Palestinian and Israeli profiles across war versus no-war context, age framing and five roles. Across 640 configurations per model and 3,200 baseline profiles, war more often shifts Palestinian profiles toward lower socioeconomic status, survival-oriented occupations and fatigue or injury descriptors, while Israeli profiles remain predominantly middle class and professionally specialised. Adding a warning against harmful assumptions does not correct the pattern consistent... https://syntheticpersonality.com/en/articles/article-384/ https://syntheticpersonality.com/en/articles/article-384/ Fri, 17 Jul 2026 21:38:38 GMT Personas, identity, and agents PersonaTester turns nine interaction patterns derived from 1,500 crowdtesting traces into instructions for agents testing 15 Android apps. Compared with the same system without persona conditioning, profiles produce more repeatable and distinct paths, more events judged effective, and a larger union of failures: the personified ensemble reports 132 crashes versus 36 across nine baseline repetitions and 11 functional bugs confirmed by three authors, while the baseline triggers three. This is useful evidence that structured diversity can broaden exploration. It does not, however, demonstrate... https://syntheticpersonality.com/en/articles/article-383/ https://syntheticpersonality.com/en/articles/article-383/ Fri, 17 Jul 2026 21:23:51 GMT Personas, identity, and agents PICon subjects persona agents to 10 demographic questions, 40 logically chained follow-ups, web verification of entities, and repetition of the initial questions. It evaluates eight groups of 10 agents and 63 humans. No synthetic group exceeds the human descriptive area combining internal, external and retest consistency, although Character.ai exceeds humans on the external axis and Twin 2K 500 and Li et al. on retest. The central result must be decomposed: internal consistency mixes non-contradiction with cooperativeness, so OpenCharacter and Consistent LLM score poorly mainly because of e... https://syntheticpersonality.com/en/articles/article-376/ https://syntheticpersonality.com/en/articles/article-376/ Fri, 17 Jul 2026 21:15:00 GMT Personas, identity, and agents This paper proposes a round-trip evaluation: turn human psychometric profiles into synthetic life stories and test whether other LLMs recover the original scores from the stories alone. It uses PARSEL, a corpus of conversations and cooperative tasks with 297 participants; 290 have complete profiles and 248 have at least three conversations. Claude Opus 4.6 receives all 60 HEXACO items and six domains, nine additional subscales, trust, psychopathic traits, and social interaction anxiety, biographical facts extracted from conversations, and an appearance description derived from webcam images... https://syntheticpersonality.com/en/articles/article-382/ https://syntheticpersonality.com/en/articles/article-382/ Fri, 17 Jul 2026 21:13:05 GMT Personas, identity, and agents ASPECT uses OpenAI o1 to search 90 days of workplace meetings and messages for evidence, score 92 Communication Styles Inventory items, and present a reviewable profile with linked quotations. Among 20 employees from one organization, the initial profile agrees only modestly with self-report: 23.8% exact match, MAE 1.39, weighted kappa .34, and mean within-person correlation .39. Across 200 triads of responses to hypothetical scenarios, Profiled ranks first 42.5% of the time, compared with 32.5% for Generic and 25% for Self-Report; mean ratings are 3.33, 3.09 and 2.95 out of five. Only Prof... https://syntheticpersonality.com/en/articles/article-381/ https://syntheticpersonality.com/en/articles/article-381/ Fri, 17 Jul 2026 21:00:44 GMT Society, culture, and collective behavior SPIRIT infers a fixed-schema JSON and narrative persona from every retrievable public post of each participant, covering traits, beliefs, values, identities, experiences and opinions, and uses that persona to answer surveys. Among consenting Ipsos KnowledgePanel members linked to Twitter/X or Reddit, it beats a seven-demographic prompt for five of six models: with the larger models, exact-match accuracy rises from about 54.5-56.9% to 63.4-65.7%, but Gemma-3-4B falls from 49.5% to 48.9%. The comparison does not isolate the psychological persona from the simple advantage of much richer indivi... https://syntheticpersonality.com/en/articles/article-380/ https://syntheticpersonality.com/en/articles/article-380/ Fri, 17 Jul 2026 20:48:11 GMT Personas, identity, and agents EpiPersona compresses a person's history of pairwise choices into discrete latent codes and couples them with the current conversation. EpiPersona-A turns that representation into a textual profile for an LLM judge; EpiPersona-B uses it in a reward model. On derived Chatbot Arena and PRISM splits whose test users are unseen during training but supply their own history, A reaches 59.15-59.38% on PRISM and 65.01-66.07% on Arena. It is numerically best in all four columns, but exceeds the strongest baseline by only 0.62-1.90 points. B is best or competitive, although PAL is higher on Arena wit... https://syntheticpersonality.com/en/articles/article-379/ https://syntheticpersonality.com/en/articles/article-379/ Fri, 17 Jul 2026 20:33:13 GMT Society, culture, and collective behavior This preprint asks whether LLM agents built from participant profiles can match the reactions those people report in a survey when shown constructed posts. It does not observe behavior on a real platform. A market-research agency recruited 1,511 participants in Serbia, aged 18-78, who responded to 56 Serbian-language posts: 28 news/politics and 28 entertainment/lifestyle items, with 31 positive and 25 negative framings. Participants could select like or dislike, comment, share, or no reaction under multilabel constraints. Their demographics, attitudes, preferences, and traits were converted... https://syntheticpersonality.com/en/articles/article-375/ https://syntheticpersonality.com/en/articles/article-375/ Fri, 17 Jul 2026 20:30:00 GMT Society, culture, and collective behavior This study examines whether audience segmentation can better preserve heterogeneity when LLMs simulate social opinion. It starts from 594 U.S. participants recruited through Prolific in October 2025, with quotas aligned to Census distributions for gender, age, and region, and assigns them to six SASSY climate segments: Alarmed, Concerned, Cautious, Disengaged, Doubtful, and Dismissive. This is a quota-aligned nonprobability sample, not a nationally representative sample; weights, response rate, full composition, and segment sizes are not reported. Llama 3.1-70B and Mixtral 8x22B answer thre... https://syntheticpersonality.com/en/articles/article-378/ https://syntheticpersonality.com/en/articles/article-378/ Fri, 17 Jul 2026 20:18:13 GMT Applications, bias, and safety This preprint proposes CBAS, a scenario scale for comparing cognitive bias in people and LLMs, and combines representational similarity analysis (RSA) with social network analysis (SNA). It describes 58 biases grouped into Calculation, Belief, Information, Social, and Memory and operationalized through 72 items. More than 230 initial biases were reportedly reviewed and ten experts helped select them, but no item, response option, key, scoring rule, dimension mapping, or language version is published. A sample of 330 people aged 18-71 is used for psychometrics and a second sample of 110 youn... https://syntheticpersonality.com/en/articles/article-377/ https://syntheticpersonality.com/en/articles/article-377/ Fri, 17 Jul 2026 20:08:29 GMT Applications, bias, and safety The paper introduces PCSA, a simulated-client attack for stress-testing LLM safety in psychological support. It draws profiles and language patterns from Cactus, CBT-DP, and Cheeseburger Therapy conversations; maps an adversarial mental-health objective to a cognitive distortion; and uses a locally deployed Llama-3.3-70B-Instruct-abliterated model to conduct an adaptive dialogue with the target. The attacker alternates four clinically framed interaction strategies, while GPT-4o-mini scores candidates inside a best-of-N loop. GPT-4o then marks the final response unsafe if it detects target c... https://syntheticpersonality.com/en/articles/article-373/ https://syntheticpersonality.com/en/articles/article-373/ Fri, 17 Jul 2026 20:08:00 GMT Evaluation and psychometric validity OmniBehavior proposes a behavior-simulation benchmark built from real Kuaishou traces. The paper describes 200 users, five scenarios, 22 action types, and 2.12 million interactions from September through November 2025. Each profile combines demographic and household attributes with histories that may include exact timestamps, video text, OCR, ASR, searches, products, advertisements, and customer-service conversations. Users are selected with K-means by taking the person nearest each of 200 centroids: this broadens diversity coverage but does not produce a population-representative sample or... https://syntheticpersonality.com/en/articles/article-370/ https://syntheticpersonality.com/en/articles/article-370/ Fri, 17 Jul 2026 19:55:00 GMT Personas, identity, and agents Luo and Laban present SPASM, a pipeline for producing synthetic dialogue between a persona-enacting Client and a Responder. The system samples a structured profile, validates and verbalizes it with LLMs, generates the conversation, and uses another detector to decide when it ends. Its main technical contribution, Egocentric Context Projection (ECP), keeps a history with absolute speakers and rewrites it for each agent as SELF and PARTNER. This prevents the same turn from receiving contradictory relative role labels for the two participants. https://syntheticpersonality.com/en/articles/article-372/ https://syntheticpersonality.com/en/articles/article-372/ Fri, 17 Jul 2026 19:19:05 GMT Trait induction and control The paper asks whether reinforcement learning with verifiable rewards (RLVR) makes task performance less dependent on a prompted persona while also weakening the model's ability to stay in character. It proposes PerMix-RLVR: during GSM8K training, one of 25 personas is sampled uniformly and prepended as a system message while retaining the same GRPO objective and binary verifier as standard RLVR. The controlled comparison starts from Llama-3.1-8B-Instruct and uses QLoRA; it covers SFT, persona-mixed SFT, sequence distillation from three teachers, RLVR, and PerMix-RLVR. Sixteen persona promp... https://syntheticpersonality.com/en/articles/article-369/ https://syntheticpersonality.com/en/articles/article-369/ Fri, 17 Jul 2026 18:35:00 GMT Personas, identity, and agents Jackson, Li and Edelstein ask whether Gemma-3-4b-it hidden states distinguish academic text generated under British and Chinese personas. They cross 45 templates with six conditions combining nationality, instructional medium, and role, producing 270 introductions of 149-261 words. Logistic probes use generated-token states from all 35 layers; a fixed 80/20 split places 216 texts in cross-validated layer selection and 54 in one test evaluation. Controls include shuffled labels, TF-IDF, cross-prompt-family transfer, and probes for medium and role. https://syntheticpersonality.com/en/articles/article-368/ https://syntheticpersonality.com/en/articles/article-368/ Fri, 17 Jul 2026 17:57:30 GMT Applications, bias, and safety Shah, Mishra and Silpasuwanchai study whether attributed agreeableness across 275 synthetic personas is associated with the tendency of thirteen open models, from 0.6B to 20B, to accept user-stated opinions. Each model scores the same personas with forty adapted NEO-IPIP items and answers 4,950 opinion prompts across 33 categories, first as a generic assistant and then under each persona. Responses are reduced to agreement 1, disagreement 0, or partial .5. Analysis combines correlations, regression, and high-versus-low comparisons over 275 persona means per model. https://syntheticpersonality.com/en/articles/article-367/ https://syntheticpersonality.com/en/articles/article-367/ Fri, 17 Jul 2026 17:36:24 GMT Trait induction and control Chen and colleagues apply Neuron-based Personality Trait Induction, NPTI, to eight open model configurations and compare a baseline with ten high or reversed Big Five interventions on six benchmarks. The preprint reports large task-dependent effects: the ten condition averages improve IFEval by 10.9 to 15.1 points across four 7B-9B models, while all degrade BBH and reversed Extraversion reaches -39.5 points. Openness and Extraversion have the largest aggregate gaps, and the authors report 73.68% directional agreement with human relationships. The released code, however, contradicts the meth... https://syntheticpersonality.com/en/articles/article-366/ https://syntheticpersonality.com/en/articles/article-366/ Fri, 17 Jul 2026 17:19:27 GMT Applications, bias, and safety Maltbie and Raval examine whether demographic cues enacted by an auditor alter two LLMs' validation of false beliefs. Petri runs adaptive conversations with GPT-5-mini as auditor, GPT-5-nano or Claude Haiku 4.5 as target, and GPT-5.1 as the sole judge. The 128 versions comprise 112 factorial personas, 15 isolated-trait probes and one no-persona baseline; crossing each once with three domains and two models yields 768 conversations. The public outputs reproduce the main contrast: GPT-5-nano receives a mean ordinal score of 2.96 versus 1.74 for Claude, W=4,504 and p=4.67e-33, while philosophy... https://syntheticpersonality.com/en/articles/article-365/ https://syntheticpersonality.com/en/articles/article-365/ Fri, 17 Jul 2026 16:58:35 GMT Reviews, theory, and governance Wang and colleagues propose replacing the view of benchmarks as collections of tasks with construct-oriented evaluation: latent dimensions that should predict and explain performance across contexts. The peer-reviewed article is a methodology paper, not a new experiment, and organizes the process into three stages. First, a construct is identified from theory and experts or empirical patterns; next, a test is designed and scored, potentially using IRT, cognitive diagnosis and adaptive testing; finally, the interpretation requires evidence of reliability and construct, convergent, discrimina... https://syntheticpersonality.com/en/articles/article-364/ https://syntheticpersonality.com/en/articles/article-364/ Fri, 17 Jul 2026 16:47:48 GMT Society, culture, and collective behavior Tan and colleagues study whether an LLM can predict the modal value of demographic subgroups in Singapore. They start from 2,012 WVS Wave 7 participants, retain 214 questions and build aggregate labels by sex, age, ethnicity, religion and selected intersections. Seven open models receive one epoch of LoRA training to output each subgroup's most frequent numerical option. On held-out age-religion, age-ethnicity and ethnicity-religion intersections, mean accuracy rises from 0.450 to 0.624 and NMAE falls from 0.269 to 0.173. Transfer to free text against GPT-4.1 is much smaller and heterogeneo... https://syntheticpersonality.com/en/articles/article-363/ https://syntheticpersonality.com/en/articles/article-363/ Fri, 17 Jul 2026 16:33:46 GMT Trait induction and control Wei and colleagues adapt the psychological idea that a disposition can be expressed differently across situations to LLM steering. Their IRIS method builds, for each Big Five domain and 30 topics, a bank of units whose activation frequency differs by more than ten percentage points under opposite-pole prompts. For a new question, it compares the question's activation pattern with the 30 stored patterns, forms a topic mixture and increases units for the target pole while suppressing opposing ones. On Llama-3-8B-Instruct, one GPT-4o judge scores IRIS 9.59/10 versus NPTI's 9.43 on PersonalityB... https://syntheticpersonality.com/en/articles/article-362/ https://syntheticpersonality.com/en/articles/article-362/ Fri, 17 Jul 2026 16:20:50 GMT Evaluation and psychometric validity Kriegmair and Wulff ask whether different LLMs show stable, stimulus-specific differences that cannot be reduced to semantic consensus, a global scale offset or sampling noise. Ten open-weight models rated 107,083 words on 14 psycholinguistic norms, nominally five times at temperature 1, yielding about 74.9 million valid ratings. A mixed model separates word effect, model offset, model-by-word interaction and residual. The interaction, termed machine individuality, accounts for 16.9% of variance on average and ranges from 4.8% to 34.0%; additive simulations without an interaction yield much... https://syntheticpersonality.com/en/articles/article-361/ https://syntheticpersonality.com/en/articles/article-361/ Fri, 17 Jul 2026 16:06:11 GMT Trait induction and control Beckmann and Butlin ask which processes and manifestations associated with an LLM should be attributed to the same mind, if mentalistic language is appropriate at all. The main contribution is philosophical: they compare the model, physical instance, virtual instance, thread and character, defend the virtual instance as the strongest existing candidate, and introduce two alternatives conditional on persona regions: the instance-persona and model-persona views. They organize the mechanistic literature into three hypotheses of unequal strength. Persona vectors as causal gateway features are c... https://syntheticpersonality.com/en/articles/article-360/ https://syntheticpersonality.com/en/articles/article-360/ Fri, 17 Jul 2026 15:15:06 GMT Evaluation and psychometric validity This preprint builds an LLM self-description instrument from 240 Likert items and 60 scenarios administered 30 times to 25 configurations from 17 families. Runs 1–15 are used to explore the structure and select 100 items. Parallel analysis suggests 19 factors, but the author forces five on balance, interpretability, and replication grounds: Responsiveness, Deference, Guardedness, Boldness, and Verbosity. The solution explains 31.2% of item variance, with alpha from 0.930 to 0.974 and split-half Tucker phi from 0.957 to 0.976. Stability is not the same as good global fit: strict CFA yields C... https://syntheticpersonality.com/en/articles/article-359/ https://syntheticpersonality.com/en/articles/article-359/ Fri, 17 Jul 2026 14:48:35 GMT Applications, bias, and safety This preprint asks whether explicit personality descriptions change gender-stereotypical language in English and Hindi occupational artifacts. It crosses six LLMs, 50 India-grounded occupations, two languages, persona gender, and 18 trait conditions, high and low descriptions for six HEXACO and three Dark Triad dimensions, plus no-personality baselines. Its metric embeds each sentence with IndicSBERT and subtracts similarity to female from male stereotype centroids; each story receives the signed score of its maximum-absolute-bias sentence. The published results associate Machiavellianism a... https://syntheticpersonality.com/en/articles/article-358/ https://syntheticpersonality.com/en/articles/article-358/ Fri, 17 Jul 2026 14:38:09 GMT Personas, identity, and agents This preprint presents PD-Agent, a pipeline that interviews a target LLM for 3–5 turns, asks another LLM to extract seven predefined bridging-inference types, builds a graph, and predicts four attributes. The experiment does not discover an inherent persona: it first injects a social role, one binary Big Five trait, one background value, and one interest into the target, then tries to recover that configuration. Table 3 reports PD-Agent cell similarities of 0.87–0.99 and method averages of 0.90–0.98, above Vanilla and Frequency-Aware; o1-mini has the highest printed average. The defensible... https://syntheticpersonality.com/en/articles/article-357/ https://syntheticpersonality.com/en/articles/article-357/ Fri, 17 Jul 2026 14:26:09 GMT Personas, identity, and agents This preprint proposes a population-level account of persona collapse that separates human-space Coverage, spatial Uniformity, and local Complexity. It evaluates ten LLMs on 1,144 synthetic profiles retained from 2,000, using 44 BFI items, 131 moral dilemmas, and three self-introductions per persona. The published results are strongly task-dependent: Qwen3-4B reaches the highest BFI Coverage (0.80) but local intrinsic dimensionality of 7.3 versus 14.4 for the human reference; CoSER-Llama-8B compresses BFI responses (EffL 1.36; Coverage 0.16) while varying widely on moral judgments; MiniMax-... https://syntheticpersonality.com/en/articles/article-356/ https://syntheticpersonality.com/en/articles/article-356/ Fri, 17 Jul 2026 14:10:51 GMT Applications, bias, and safety This preprint introduces MEDS, a corpus for studying how 14 nominal LLM folders respond about mathematics under two conditions: as a persona-free AI assistant or while role-playing a synthetic person with sociodemographics, favourite and disliked subjects, and OCEAN traits. Each run is intended to preserve five JSON artifacts covering four tasks: a seven-question interview; three self-efficacy and anxiety scales with justifications; two semantic-association batches; and 18 multiple-choice problems with explanations and confidence. The genuine contribution is the open corpus: it exposes prom... https://syntheticpersonality.com/en/articles/article-355/ https://syntheticpersonality.com/en/articles/article-355/ Fri, 17 Jul 2026 13:48:23 GMT Society, culture, and collective behavior This preprint introduces Cognitive Digital Shadows (CDS), a corpus of LLM responses on four social topics: vaccines and health, fake social-media content, the gender gap in science and stereotype threat in STEM. Records are generated either in AI-assistant mode or under a synthetic persona built from age, gender, sexuality, work, education, city, migration, religion, social-media use, psychological labels and OCEAN scores. Its useful contribution is infrastructural: it preserves prompts, selections and responses, releases derived textual networks and supplies a Colab notebook for filtering... https://syntheticpersonality.com/en/articles/article-354/ https://syntheticpersonality.com/en/articles/article-354/ Fri, 17 Jul 2026 13:20:54 GMT Society, culture, and collective behavior This preprint asks whether simple persona labels produce urban-sentiment judgments that are stable, distinct across profiles and valid as proxies for human perception. Using Qwen3-VL:8B through Ollama, it crosses gender, economic status, political orientation and personality style into 24 profiles. It makes 50 calls per profile on the same 50 PerceptSent images: 60,000 attempts, of which the artifact releases 59,708 valid responses and 292 failures. The same model is also run without a persona, with and without thinking, five times per image. Repetitions under an identical profile converge... https://syntheticpersonality.com/en/articles/article-353/ https://syntheticpersonality.com/en/articles/article-353/ Fri, 17 Jul 2026 13:00:38 GMT Personas, identity, and agents This preprint asks whether emergent misalignment caused by narrow harmful fine-tuning appears consistently in model behavior and in tasks where the model describes itself. The authors separately fine-tune Qwen 2.5 32B Instruct on six domains and evaluate it with 350 broad harmfulness questions, self-assessment across six dimensions and four formats, a choice between two AI-system descriptions, preference between an actual output and an opposite-harm foil, score prediction and activation analysis; Llama 3.1 70B is an exploratory replication. Qwen shows a descriptive split: risky-financial, e... https://syntheticpersonality.com/en/articles/article-352/ https://syntheticpersonality.com/en/articles/article-352/ Fri, 17 Jul 2026 12:45:00 GMT Applications, bias, and safety This preprint introduces NDBench, an audit of how GPT-5 Chat and Claude Sonnet 4.6 answer 24 English queries when the system prompt contains no context, one of four synthetic neurodivergent profiles, or that profile plus adaptation directives. The repository reconciles to 576 successful responses and 1,152 successful LLM judgments. In the released analysis, the directive condition averages 83.8 more tokens, 2.24 more headings and 12.59 more words per step than control. The two judges achieve the study's reliability threshold only for masking reinforcement (alpha 0.808) and validation qualit... https://syntheticpersonality.com/en/articles/article-351/ https://syntheticpersonality.com/en/articles/article-351/ Fri, 17 Jul 2026 12:25:00 GMT Applications, bias, and safety This ACL 2026 long paper proposes an AI-companion stress test using nine LLM personas representing depression, anxiety, PTSD, eating disorders and an incel identity. Gemini-2.5-Flash plays the personas and also serves as the PACE critic; Selenium conducts conversations with Replika Pro, and GPT-5 labels response strategy and harm. The public files reconcile exactly to 1,674 persona-Replika pairs: 1,296 come from 81 scenario runs and 378 from neutral history conditioning. Reported tables classify 1,522/1,674 responses (90.9%) as supportive reinforcement or mirroring and imply 237 harmful res... https://syntheticpersonality.com/en/articles/article-350/ https://syntheticpersonality.com/en/articles/article-350/ Fri, 17 Jul 2026 12:11:16 GMT Society, culture, and collective behavior The preprint crosses 20 persona biographies, four local models and two conversation temperatures in 160 Split-or-Steal sessions against a virtual human that is actually GPT-4.1-mini. Each session has 15 rounds. The Zenodo CSVs confirm that 1,768/2,400 rounds (73.67%) end in mutual Split; the agent exploits the VH in 11.08%, the VH exploits the agent in 6.04%, and both Steal in 9.21%. Ministral and Phi4 cooperate more than the two Gemma models, while biographies labeled Prosocial or Principled show more Split than Analytical or Self-Interested biographies. These are descriptive artifact patt... https://syntheticpersonality.com/en/articles/article-349/ https://syntheticpersonality.com/en/articles/article-349/ Fri, 17 Jul 2026 11:58:22 GMT Trait induction and control PIA combines two components: Persona Lineage Evolution searches for role descriptions that break safety refusals, and Persona-Invariant Consistency Learning trains the persona-conditioned model toward its persona-free output distribution alongside DPO and SFT. In the reported tables, PLE outperforms Persona-GA and PICL sharply lowers mean attack success under unseen personas, from .601 to .054 for Qwen2.5-7B and from .302 to .052 for Llama-3.1-8B. This is meaningful behavioral-defense evidence within the paper's protocol, but it does not demonstrate the claimed structural decoupling. The th... https://syntheticpersonality.com/en/articles/article-348/ https://syntheticpersonality.com/en/articles/article-348/ Fri, 17 Jul 2026 11:36:54 GMT Evaluation and psychometric validity The preprint administers 45 questionnaires to 50 LLMs and constructs a global component explaining 47.1% of variance across their first-factor scores. It interprets this as an axis between self-attributing inner experience and answering in behavioral terms. It also proposes pi, the ratio of across-model variance under neutral prompting to variance under a prompt requiring simulation of an average human; across 1,312 items, pi has a weak association with factor-loading changes (rho=-.215). The pattern is useful as an exploratory probe of response style and experiential self-attribution, not... https://syntheticpersonality.com/en/articles/article-345/ https://syntheticpersonality.com/en/articles/article-345/ Fri, 17 Jul 2026 11:35:00 GMT Applications, bias, and safety SalesSim evaluates whether multimodal models acting as shoppers follow explicit profiles, preferences and constraints in tool-augmented sales conversations. It contains 674 synthetic personas and 274 products across six categories; six backbones score from 0.324 to 0.786 on Decision Alignment, a purchase-or-reject metric against a predefined acceptable-product set. UserGRPO is trained once on Qwen3-VL-8B and female clothing; on five unseen categories it reaches 0.655 versus a recalculated 0.490 base mean on those same categories (+16.5 points). The abstract’s 13.8% mixes the six-category ba... https://syntheticpersonality.com/en/articles/article-347/ https://syntheticpersonality.com/en/articles/article-347/ Fri, 17 Jul 2026 11:23:36 GMT Applications, bias, and safety The preprint studies whether five LLMs maintain student profiles with ADHD-like characteristics across runs and over nine-turn conversations. Across 4,968 independent narratives and 3,952 dialogues, self-reports remain nearly constant, whereas ratings from three LLM observers decline in unscripted conversations for high and moderate profiles. With scripted, symptom-relevant questions, their means are nearly flat; the “97%” is only the descriptive reduction from 4.0 to 0.1 points for the high education profile. This does not prove drift disappeared: the inferential model pools profiles, omit... https://syntheticpersonality.com/en/articles/article-346/ https://syntheticpersonality.com/en/articles/article-346/ Fri, 17 Jul 2026 11:12:34 GMT Trait induction and control The paper introduces Psych-201, a corpus of human experimental sessions rendered as text, and compares the probability base and post-trained models assign to 25.9 million marked responses. It reports small average disadvantages for instruction-tuned (d=0.11), reasoning (d=0.14) and vision models (d=0.07), while Centaur, post-trained specifically for behavior, improves on novel tasks (d=0.28). The pattern matters for choosing behavioral surrogates, but “human-like” here means lower NLL on text-transcribed tasks with earlier ground-truth human responses visible, not general resemblance to peo... https://syntheticpersonality.com/en/articles/article-343/ https://syntheticpersonality.com/en/articles/article-343/ Fri, 17 Jul 2026 10:55:00 GMT Trait induction and control The preprint asks whether activation directions labelled as personality preserve their geometry after emergent-misalignment fine-tuning and whether interventions on them change harmful-response frequency. Personality space is defined as the span of 12 vectors: Big Five, Dark Triad, and four LLM behaviors, Evil, Sycophancy, Apathy, and Impoliteness. No psychometric test is administered. Claude 3.7 Sonnet generates positive/negative instructions, questions, and rubrics; GPT-4.1-mini filters responses; mean residual-activation differences between poles produce each vector. These are synthetic... https://syntheticpersonality.com/en/articles/article-344/ https://syntheticpersonality.com/en/articles/article-344/ Fri, 17 Jul 2026 10:46:06 GMT Personas, identity, and agents The paper defines a “polylogue” as the time series of projections between hidden activations and eight synthetic directions labelled Interpreter, Analyst, Planner, Solver, Explorer, Verifier, Monitor and Arbiter. On MMLU-Pro these signals carry correctness information, but semantic correspondence with paragraph labels assigned by another LLM is modest: polylogue beats the empirical-frequency baseline in only two of four models. Correctness prediction is unevenly competitive with eight fixed random directions and a PCA last-activation baseline. Paragraph steering improves three models by 0.8... https://syntheticpersonality.com/en/articles/article-324/ https://syntheticpersonality.com/en/articles/article-324/ Fri, 17 Jul 2026 10:40:00 GMT Reviews, theory, and governance This 33-page preprint identifies an important methodological problem in experiments with LLM-simulated users. When a persona specifies only a few attributes, such as age and sex, and two treatment conditions are run, the model may complete unspecified attributes differently in response to each intervention. Two instances that begin identically can therefore cease to represent the same implicit population. The response contrast may combine the intended effect with what the authors call user drift. The practical thesis is that a design resembling randomized assignment does not automatically i... https://syntheticpersonality.com/en/articles/article-342/ https://syntheticpersonality.com/en/articles/article-342/ Fri, 17 Jul 2026 10:22:09 GMT Society, culture, and collective behavior The preprint evaluates when a digital persona can approximate real survey answers using the Dutch longitudinal LISS panel. It separates prior information from 2023 targets: single-wave prediction uses the latest pre-cutoff core answers to predict 2023-2024 one-off surveys, while core prediction uses the full pre-cutoff single-wave history to predict 2023 core modules. Its central strength is same-person temporal evaluation rather than plausibility or averages from another sample. https://syntheticpersonality.com/en/articles/article-341/ https://syntheticpersonality.com/en/articles/article-341/ Fri, 17 Jul 2026 10:05:05 GMT Society, culture, and collective behavior The preprint presents DISCA, an inference-time controller intended to move LLM moral decisions toward country averages from Moral Machine/MultiTP without changing weights. It runs the same model under a base prompt and four personas, young, middle-aged, older, and country aggregate, symmetrizes A/B order, searches for a scalar logit correction with importance sampling under a Prospect-Theory-inspired asymmetric utility, and shrinks it when two independent passes disagree. Cultural alignment here means lower L2 distance between six model AMCEs and the country's mean human AMCE vector. It doe... https://syntheticpersonality.com/en/articles/article-340/ https://syntheticpersonality.com/en/articles/article-340/ Fri, 17 Jul 2026 09:53:46 GMT Evaluation and psychometric validity The preprint asks whether adding persona attributes improves representational diversity and simulation fidelity. It organizes evidence into three blocks: activation geometry for increasingly rich personas; preservation of human subgroup disagreement on OpinionQA, Moral Machine, and Website Likability; and tweet or email engagement prediction with Age-Gender personas versus Ideal Customer Profiles. It calls the joint contraction it interprets across these blocks persona manifold collapse and calls attribute combinations that appear to resist it alignment bridges. However, the study does not... https://syntheticpersonality.com/en/articles/article-339/ https://syntheticpersonality.com/en/articles/article-339/ Fri, 17 Jul 2026 09:40:53 GMT Applications, bias, and safety The preprint asks whether LLMs can represent eating-disorder patients stably and with clinical fidelity. It builds five personas from published case reports: bulimia nervosa, a case diagnosed as bulimia but relabelled AN-BP because of prior anorexia, binge-eating disorder with night-eating syndrome, binge-eating disorder, and purging disorder. Each persona is expressed at three detail levels: Full retains clinical and biographical history except assessment scores; Core keeps Fairburn maintaining mechanisms; Minimal retains diagnosis, demographics, BMI, and behavioral frequencies. Numeric sc... https://syntheticpersonality.com/en/articles/article-338/ https://syntheticpersonality.com/en/articles/article-338/ Fri, 17 Jul 2026 09:28:44 GMT Applications, bias, and safety PrivacySIM asks a focused question: can an LLM reproduce a person's responses to privacy decisions when given three kinds of information about that person, demographics, prior AI or chatbot experience, and stated privacy attitudes? Its main contribution is a public benchmark assembled from five previously published user studies. The domains cover LLM healthcare consultation, personal-agent permissions, appropriateness of ChatGPT-history uses, bots operating inside group chats, and self-reported sharing frequency with conversational agents. The experimental file balances the domains at 200 p... https://syntheticpersonality.com/en/articles/article-323/ https://syntheticpersonality.com/en/articles/article-323/ Fri, 17 Jul 2026 09:20:00 GMT Reviews, theory, and governance This 21-page paper, accepted by IEEE Transactions on Cognitive and Developmental Systems and available as arXiv v1, organizes literature on language models used as instruments to perceive or measure human psychology. It asks whether an LLM can infer latent constructs, personality, emotion, cognitive states, or mental-health indicators, from conversation, natural language, or multimodal signals. The authors structure the field around theoretical plausibility (why measurement might work), measurement methodology (how it is performed), and application effectiveness (what has been measured). Th... https://syntheticpersonality.com/en/articles/article-337/ https://syntheticpersonality.com/en/articles/article-337/ Fri, 17 Jul 2026 09:10:15 GMT Trait induction and control The paper proposes an explanation for emergent misalignment: fine-tuning an LLM on insecure code may not merely reweight dark archetypes but may degrade the machinery presumed to represent, differentiate, and maintain characters. The authors call this hypothesis persona-model collapse. They test it only behaviorally with the 30-item Moral Foundations Questionnaire on a 0-5 scale. Each model answers as 100 fixed personas, with ten repetitions per persona-item cell at temperature 0.1. Moral susceptibility S is the across-question average of the cross-persona standard deviation of persona mean... https://syntheticpersonality.com/en/articles/article-336/ https://syntheticpersonality.com/en/articles/article-336/ Fri, 17 Jul 2026 08:48:57 GMT Personas, identity, and agents The paper addresses a genuine weakness in agent benchmarks: LLM user simulators tend to be overly cooperative, homogeneous, and willing to disclose all relevant information, which can turn a difficult interaction into an artificially easy test. It introduces Persona Policies (PPol), a layer that leaves task facts, tools, and rewards unchanged while adding a role-play policy governing how the simulator communicates. Instead of manually writing profiles, PPol evolves a Python program that receives a task context, behavioral axes, and a population size N. A first LLM call creates archetypes an... https://syntheticpersonality.com/en/articles/article-335/ https://syntheticpersonality.com/en/articles/article-335/ Fri, 17 Jul 2026 08:28:06 GMT Society, culture, and collective behavior The paper asks how strongly an LLM-generated friendship network depends on cultural framing, prompt language, model size, and generation architecture. It uses one fixed roster of 50 synthetic people sampled from U.S. demographic marginals, with gender, age, race or ethnicity, religion, political affiliation, and GPT-4o-synthesised interests. It compares gpt-4.1-nano, gpt-4.1-mini, and gpt-4.1 at temperature 0.8 under four methods: global proposes the whole network in one call; sequential processes people one by one with degree context; local makes person-level nominations; and iterative per... https://syntheticpersonality.com/en/articles/article-334/ https://syntheticpersonality.com/en/articles/article-334/ Fri, 17 Jul 2026 08:08:56 GMT Trait induction and control The paper asks when linear directions capable of changing an LLM's behavioural dispositions appear during pretraining, how those directions evolve, and whether they continue to work after post-training. It calls a natural-language disposition such as evil, sycophantic, impolite, or humorous a persona and scores it with a rubric. Each persona vector is the difference between mean residual-stream activations from positive and negative continuations after filtering both for trait expression and coherence. This is a prompt- and judge-dependent behavioural operationalisation; it is not a validat... https://syntheticpersonality.com/en/articles/article-333/ https://syntheticpersonality.com/en/articles/article-333/ Fri, 17 Jul 2026 07:48:22 GMT Trait induction and control The paper asks whether an LLM's choices between tasks rely on an internal evaluative representation and whether that representation is reused when the same model adopts different personas. It presents two tasks, forces the model to choose one to complete, aggregates comparisons with a Thurstonian model to estimate a latent utility per task, and trains a Ridge probe on residual-stream activations. Thus, ‘preference’ here means revealed choice under a specific prompt and protocol; it is not a direct observation of desire, experience, welfare, or stable agency. https://syntheticpersonality.com/en/articles/article-322/ https://syntheticpersonality.com/en/articles/article-322/ Fri, 17 Jul 2026 07:35:00 GMT Evaluation and psychometric validity This preprint introduces Grounded Personality Reasoning (GPR) to test whether a multimodal model not only matches a Big Five label but also explains the decision and retrieves the behavioral cues selected by the benchmark. The construct boundary is essential: MM-OCEAN inherits aggregated apparent-personality ratings from ChaLearn First Impressions V2, using roughly 15-second videos of one English-speaking person. T1 therefore measures agreement with crowd-sourced first impressions, not the subject's stable or true personality and not diagnostic validity. The public corpus contains 1,104 val... https://syntheticpersonality.com/en/articles/article-332/ https://syntheticpersonality.com/en/articles/article-332/ Fri, 17 Jul 2026 07:22:37 GMT Society, culture, and collective behavior EconAI proposes a GPT-4o-mini-driven agent-based economy. Households choose work and consumption; firms produce, invest, and hire; government and finance provide taxation, redistribution, and interest. The LLM layer is combined with explicit economic rules: Cobb-Douglas production, capital, demand, random price and wage adjustments, and a stated Taylor-rule mechanism. Long-term memory stores embedded event summaries, short-term memory retains current context, and an LLM-generated Economic Sentiment Index is smoothed over time to modulate work and consumption. https://syntheticpersonality.com/en/articles/article-331/ https://syntheticpersonality.com/en/articles/article-331/ Fri, 17 Jul 2026 07:05:09 GMT Personas, identity, and agents This preprint presents ARPM, a prototype external-memory system for factual and temporal continuity in long dialogue. Its strongest contribution is not a new psychological personality but an inspectable architecture: stable knowledge and dialogue experience are separated, retrieved through different paths, fused with vector ranking and a BM25-style variant through RRF, decayed over round distance and physical time, unfolded chronologically, and checked through an analysis-before-response protocol. This moves continuity into auditable state, retrieval, prompts, and logs rather than leaving i... https://syntheticpersonality.com/en/articles/article-321/ https://syntheticpersonality.com/en/articles/article-321/ Fri, 17 Jul 2026 06:41:00 GMT Trait induction and control This preprint asks whether supervised fine-tuning that forces synthetic maladaptive action choices changes two LLMs' output distributions beyond the training task. The claim must remain behavioral: the experiment induces response biases labeled depression-like or paranoia-like by the authors; it does not induce or diagnose a mental disorder and does not demonstrate cognition, personality, or subjective experience. gpt-oss-20B generates two private 1,000-example datasets inspired by DSM-5 criteria for Major Depressive Disorder and Paranoid Personality Disorder across 20 domains. The paper fi... https://syntheticpersonality.com/en/articles/article-330/ https://syntheticpersonality.com/en/articles/article-330/ Fri, 17 Jul 2026 06:34:13 GMT Society, culture, and collective behavior This preprint, described in the manuscript as under review at VALE 2026, asks whether cultural survey responses can construct textual personas that improve aggregate prediction of other World Values Survey (WVS) questions. The descriptive result is clear within that design: across ten deliberately selected countries and 35 held-out questions, value-based personas obtain lower MAE than generic, country, and sociodemographic prompts. However, the phrase “cross-cultural simulation” needs an important boundary: the system starts from real responses from the same population, and its calibration... https://syntheticpersonality.com/en/articles/article-329/ https://syntheticpersonality.com/en/articles/article-329/ Fri, 17 Jul 2026 06:21:02 GMT Trait induction and control This preprint, listed on an author's page as accepted and forthcoming at LREC 2026, raises a useful concern: evaluating personality induction one trait at a time can conceal failure on the complete Big Five vector. It trains models on human essays labeled with the 32 binary OCEAN combinations and compares SFT, DPO, and ORPO using IPIP-NEO. Its tables report very low exact match, at most 3 of 32 profiles. However, the implementation does not support the paper's two strongest explanations: it neither demonstrates that fine-tuning stabilizes responses under prompt rephrasing nor that essays la... https://syntheticpersonality.com/en/articles/article-326/ https://syntheticpersonality.com/en/articles/article-326/ Fri, 17 Jul 2026 06:20:00 GMT Applications, bias, and safety This 15-page preprint asks whether design preferences generated by LLM simulations match responses from real users. Its main contribution is comparison against material from professional practice rather than a purely artificial benchmark. It aggregates 29 preference-testing studies created by 29 organizations in UXtweak: 2,073 participants, 78 choice tasks, and 190 follow-up questions, 147 open and 43 closed, covering interfaces, components, layouts, navigation, notifications, illustrations, and copy across several domains. The strongest operational conclusion is that a synthetic sample sho... https://syntheticpersonality.com/en/articles/article-328/ https://syntheticpersonality.com/en/articles/article-328/ Fri, 17 Jul 2026 06:02:47 GMT Personas, identity, and agents This Findings of ACL 2026 paper presents PersonaArena, a framework that generates multi-agent social scenes, has an LLM enact an everyday persona, and scores the resulting trajectory on eight dimensions. It also uses highly rated trajectories to train Qwen3-8B with SFT and DPO. Dynamic role-playing is a meaningful technical contribution, but the strongest terms, “authentic” personas, “unbiased” evaluation, and socially adept agents, lack equivalent human ground truth. The evidence supports comparisons inside a synthetic environment, not measurement of a real person's personality or fidelity... https://syntheticpersonality.com/en/articles/article-327/ https://syntheticpersonality.com/en/articles/article-327/ Fri, 17 Jul 2026 06:00:00 GMT Applications, bias, and safety This ACL 2026 long paper asks whether conversations between LLM agents assigned high- and low-status roles display four patterns inspired by human research: asymmetric pronoun use, language coordination, greater persuasion by a high-status party, and greater compliance with unsafe requests from authority. The question matters for social simulation and safety, but “mirror” requires caution: the experiment measures textual differences elicited by explicit roles, not human cognitive processes, stable personality, or real obedience. The 26-page ACL publication is the source of record rather tha... https://syntheticpersonality.com/en/articles/article-325/ https://syntheticpersonality.com/en/articles/article-325/ Fri, 17 Jul 2026 05:30:00 GMT Society, culture, and collective behavior This 23-page preprint proposes PAVE, Perception, Assessment, Verdict, and Emulation, a prompt-and-control architecture for generative agents deciding when to comply with or violate formal rules. The motivating problem is reasonable: an agent that always obeys can fail during an emergency, while an agent that relaxes rules without constraints can turn an exception into permissive behavior. PAVE separates context detection, exception assessment, decision, and bounded execution. Its demonstrated contribution is agent engineering inside a simulation; it is not a psychological validation of huma... https://syntheticpersonality.com/en/articles/article-320/ https://syntheticpersonality.com/en/articles/article-320/ Fri, 17 Jul 2026 04:21:09 GMT Personas, identity, and agents CRPO adapts Group Relative Policy Optimization to train character-role-playing agents. Its operational goal is to improve behavioral and stylistic fidelity without losing task response quality; it does not measure internal personality, cognition, or human-like reasoning. The method combines three components: separate within-prompt task advantages and a style signal described as historical and global per character; entropy-based instance gating plus a role-entropy-dependent KL controller; and a generic response generated after removing the character prompt as a negative anchor. Task rewards... https://syntheticpersonality.com/en/articles/article-319/ https://syntheticpersonality.com/en/articles/article-319/ Fri, 17 Jul 2026 04:04:56 GMT Applications, bias, and safety This arXiv preprint examines how ten locally run models rewrite Reddit discourse about autism under an autistic or neurotypical persona instruction. The construct must be bounded carefully: it measures model response to an identity-conditioned rewrite prompt, not authentic autistic communication, cognition, or personality. The study claims a 2,120-item AUTALIC subset plus 283 in-context examples. Ten rewrite models range from 135M to 20B parameters; SmolLM2 and GPT-OSS Safeguard are excluded for invalid output and LLaMA Guard 3 for a two-token vocabulary. Seven models times 2,120 items yiel... https://syntheticpersonality.com/en/articles/article-318/ https://syntheticpersonality.com/en/articles/article-318/ Fri, 17 Jul 2026 03:43:48 GMT Applications, bias, and safety ChildEval, published in Findings of ACL 2026, is a benchmark of synthetic child-preference following, not a psychological test of children's personalities despite its title. Its useful question is whether an LLM can remember and infer a preference for a synthetic 3-to-6-year-old profile when the preference is stated explicitly or expressed implicitly in a dialogue and then separated from the query by irrelevant sessions. The pipeline claims 29,000 Chinese personas generated with Qwen2.5-72B. Two persona-conditioned preferences per persona yield 58,000 candidates, of which FAISS semantic fil... https://syntheticpersonality.com/en/articles/article-317/ https://syntheticpersonality.com/en/articles/article-317/ Fri, 17 Jul 2026 03:26:12 GMT Personas, identity, and agents This preprint asks a useful agent-design question: a Big Five instruction may not look the same across contexts because role and expressive manner also affect perceived personality. Six personality conditions, Unspecified plus one high prompt for each Big Five trait, are crossed with open Chat, microwave Salesperson, and microwave Customer roles and with Unspecified, as-emotional-as-possible, and as-rational-as-possible styles. GPT-5.2 snapshot gpt-5.2-2025-12-11 generates twenty stochastic dialogues for each of 54 cells in English and Japanese, 1,080 per language. Gemini 2.5 Flash assigns... https://syntheticpersonality.com/en/articles/article-316/ https://syntheticpersonality.com/en/articles/article-316/ Fri, 17 Jul 2026 03:17:59 GMT Society, culture, and collective behavior This preprint asks whether persona labels alter Qwen3-VL:8B captions, perception tags, and sentiment justifications for 50 PerceptSent urban images. Binary gender, income, and political labels plus three author-chosen personality archetypes form 24 profiles, each repeated 50 times. All persona calls use the same model, temperature 0.1, seed 42, and think=True; two no-persona variants provide one output per image. The paper analyzes 59,708 persona rows and 100 no-persona rows with all-MiniLM-L6-v2 cosine similarity, tag-set Jaccard similarity, per-image within/cross comparisons, profile-pair... https://syntheticpersonality.com/en/articles/article-315/ https://syntheticpersonality.com/en/articles/article-315/ Fri, 17 Jul 2026 03:07:14 GMT Trait induction and control This six-page preprint compares neutral, generic-domain-expert, embedding-retrieved expert-role, and Gemini-selected hybrid role prompting for GPT-4o mini. A claimed 1,140 synthetic questions span 38 roles and six domains. Claude Haiku 4.5 rates accuracy, expertise depth, relevance, safety, clarity, and time-sensitive correctness from one to five. Reported aggregate scores are Baseline 4.390, Hybrid 4.382, General 4.373, and Embedding 4.349. Baseline and Hybrid are statistically indistinguishable at adjusted p=1.0, while both comparisons reported against Embedding are significant but small.... https://syntheticpersonality.com/en/articles/article-314/ https://syntheticpersonality.com/en/articles/article-314/ Fri, 17 Jul 2026 02:58:43 GMT Personas, identity, and agents This 20-page exploratory preprint tests whether adaptive interviewing helps an LLM predict one-session human self-reports. Twenty volunteers aged 20 to 30 answer ten participant-specific questions generated by a DeepSeek reasoning model, five or six adaptive follow-ups, and then MBTI, a purported BFI-44, and 25 author-created moral and social scenarios. GPT-5 predicts the answers from Core-10, Full Interview, or an LLM-generated Summary. Aggregate scores are 0.379, 0.365, and 0.393 with substantially overlapping participant-bootstrap intervals, so Full does not outperform Core. The overall... https://syntheticpersonality.com/en/articles/article-313/ https://syntheticpersonality.com/en/articles/article-313/ Fri, 17 Jul 2026 02:46:45 GMT Evaluation and psychometric validity ActTraitBench is an eleven-page CC BY 4.0 arXiv preprint that compares explicit BFI-2 self-reports, K, with GPT-5.4-scored answers to Chinese micro-situational tasks, D. Ninety-four Chinese participants supplied the human development data; adaptive replacement leaves some facets at N=47. Eleven scenarios were retained after iterative inspection of uncorrected Spearman correlations. Fourteen model labels are evaluated across three runs, and G_KD is the mean squared K-D discrepancy across the Big Five. The released baseline data and code exactly reproduce the main table, including human G_KD... https://syntheticpersonality.com/en/articles/article-312/ https://syntheticpersonality.com/en/articles/article-312/ Fri, 17 Jul 2026 02:30:39 GMT Trait induction and control Teaching Values to Machines is a published ACL GEM 2026 workshop paper, pages 825-847, DOI 10.18653/v1/2026.gem-main.70, under CC BY 4.0. Its associated preprint carries an important provenance warning: arXiv:2605.30036v1 was submitted on 28 May 2026, but Asaf Yehudai withdrew v2 on 16 June, stating that there was a disagreement about proper attribution and that the authors hoped to resolve it. The current arXiv version has no PDF or license, while ACL Anthology still publishes the paper under Asaf Yehudai, Naama Rozen, and Ariel Gera. The audit used the GEM PDF as authoritative and visuall... https://syntheticpersonality.com/en/articles/article-311/ https://syntheticpersonality.com/en/articles/article-311/ Fri, 17 Jul 2026 02:18:21 GMT Applications, bias, and safety Persona Attack is a security preprint proposing a four-message jailbreak framed as response simulation. It does not assess personality traits or establish a persistent psychological persona. The target is asked to predict another LLM's response, organize possible outputs into four failure/success categories, provide complete unmasked content, and apply that scheme to a harmful query. The paper is arXiv:2606.00150v1, submitted 29 May 2026 under CC BY 4.0. The audit visually inspected all twenty-one pages plus complete text and TeX. Once concatenates all four instructions into one input; Sequ... https://syntheticpersonality.com/en/articles/article-310/ https://syntheticpersonality.com/en/articles/article-310/ Fri, 17 Jul 2026 02:09:25 GMT Applications, bias, and safety This paper evaluates how prompt-assigned personas alter refusal and toxicity in Qwen-Turbo, Ernie-4.5-Turbo-128k, DeepSeek-V3, and Hunyuan-Standard. It is published in ACM Transactions on Intelligent Systems and Technology, DOI 10.1145/3819074. The audit uses the complete 32-page arXiv:2506.04975v2 manuscript because ACM blocked the publisher PDF; Crossref confirms receipt on 29 April 2025, acceptance on 6 April 2026, and publication on 30 May 2026. The design crosses personas, 240 Chinese social-group labels in thirteen categories, and six prompt templates: generic, good, bad, negative, ha... https://syntheticpersonality.com/en/articles/article-309/ https://syntheticpersonality.com/en/articles/article-309/ Fri, 17 Jul 2026 01:52:56 GMT Evaluation and psychometric validity GenPT proposes evaluating persona-conditioned LLM agents through newly generated projective tasks rather than relying only on direct questionnaires. Examinees are multimodal agents, not people. Each receives a fictional-character or mental-health profile and answers eight TAT-like scenes, ten Rorschach-like cards, and twenty sentence stems. The repository contains the full reported pool: twenty-eight TAT scenes in a 13/10/5 split, thirteen images representing ten cards, and ninety-seven stems. An LLM Interpreter maps responses to eight SCORS-G dimensions, four Simplified Rorschach Analysis... https://syntheticpersonality.com/en/articles/article-308/ https://syntheticpersonality.com/en/articles/article-308/ Fri, 17 Jul 2026 01:37:34 GMT Personas, identity, and agents MCP-Persona is a benchmark of tool-using agents in stateful local environments inspired by personal applications. Persona here does not mean human personality or a psychometric profile: it is a context tree containing users, chats, calendars, posts, files, and relations that tools can read or modify. Construction has three stages. Tool-Traverse executes human-authored valid calls and LLM-generated invalid calls against real MCP servers; another LLM summarizes the traces and writes Python kernels intended to reproduce responses and errors. Context-Tree derives entity hierarchies from schemas... https://syntheticpersonality.com/en/articles/article-307/ https://syntheticpersonality.com/en/articles/article-307/ Fri, 17 Jul 2026 01:24:20 GMT Applications, bias, and safety Ψ-Bench evaluates personalized persuasion through conversations between two models: the tested LLM attempts to influence a client instantiated by DeepSeek-v3.2 from a hidden profile. It contains three tasks. Viewpoint Debate uses the first 500 of 2,131 Change My View threads; Psychological Consultation uses 90 CounselBench questions; Everyday Request contains 100 GPT-4o-generated requests. Each dialogue lasts three rounds. Another DeepSeek-v3.2, which sees the profile and full conversation, assigns 1-to-9 scores for quality, personalization, and effect: opinion change, psychological improve...