{
  "id": 820617,
  "title": "Experts warn ChatGPT isn't just predicting words anymore — It's now predicting human thoughts",
  "url": "https://urgent.news/2026/08/13/experts-warn-chatgpt-isnt-just-predicting-words-anymore-its-now",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-13T23:25:00.000Z",
  "source": {
    "name": "TechRadar",
    "slug": "techradar",
    "url": "https://www.techradar.com/pro/chatgpt-isnt-just-predicting-words-anymore-its-now-predicting-human-thoughts"
  },
  "original_language": "en",
  "account": "Recent research published in iScience suggests that the popular AI language model GPT-4 can anticipate collective human responses to personality questions before any individual participant even attempts to answer. Led by researchers Rotem Monsa, Aviv Zohar, and Shahar Arzy from the Hebrew University and Hadassah Medical School, the study tested whether GPT-4 could infer patterns in how people typically respond to such queries based solely on the language it was trained on, rather than generating its own questionnaire text.\n\nThe team constructed two sets of personality questionnaires, one drawing heavily from the widely accepted DSM-5 diagnostic manual for personality disorders, and the other based on an astrology textbook that assigns personality traits to each of the twelve zodiac signs. These questionnaires were presented to 600 participants alongside the Big Five Inventory, a highly validated tool for measuring core personality traits.\n\nRemarkably, GPT-4's predictions of average responses to the DSM-5-based questions correlated with the actual participant averages at a 0.71 level of consistency, while its predictions for astrology-based questions achieved a 0.85 level. These strong correlations indicate that GPT-4 successfully captured population-level response tendencies embedded within the language of both texts. However, the astrology questionnaire demonstrated notably weaker internal consistency compared to the DSM-derived one, suggesting the model may struggle to discern meaningful patterns from less scientifically grounded material.\n\nBeyond predicting overall tendencies, GPT-4 also demonstrated the ability to estimate relationships between individual questions. For both DSM-5 and astrology questionnaires, the model produced correlations of 0.74 and 0.69, respectively. This finding implies the model may be able to discern patterns in how different aspects of personality are interrelated when learned from extensive textual data.\n\nHowever, the researchers caution that GPT-4's ability to predict aggregate responses does not necessarily mean it can accurately measure individual differences in personality. The study found important differences in the statistical structures underlying the questionnaires, with the Big Five Inventory most closely matching expectations, while the DSM and astrology questionnaires showed weaker confirmatory factor-model fits. Additionally, astrology-derived items failed to preserve their intended elemental structure, instead forming broader patterns resembling existing personality dimensions.\n\nDespite these limitations, the study provides evidence that GPT-4 can anticipate collective human response patterns from textual information with measurable accuracy. Lead author Monsa notes that while LLMs like GPT-4 were not explicitly trained on psychology or personality theories, they may have learned the structure of human personality as a natural byproduct of being trained on vast amounts of language data.",
  "summary": "Researchers found GPT-4 generates convincing personality questions from an oven manual, though only clinical questionnaires kept a coherent statistical structure.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}