Can an AI read your ordinary writing and work out your personality, mood or motives, without ever being told?
Can large language models extract psychological traits directly from natural language text?
This explores whether an LLM can read ordinary written text and infer psychological characteristics from it, such as personality, mood or motives, instead of being told them.
This explores whether an LLM can read ordinary text and infer psychological traits from it. The corpus points to yes for the ingredients, but none of these notes runs the full test of handing a model someone's writing and checking its trait readout against ground truth. The evidence is indirect, and it comes from several directions.
The closest result runs in reverse. Models were given Big Five personality scores and asked to write plain-language personality summaries. The summaries carried more than the numbers did: from them alone, models predicted nine other psychological scales zero-shot, and summary plus scores beat either alone (Can language summaries unlock hidden psychological patterns?). So language is a rich enough container for psychology that a model can read hidden structure out of it. The corpus also shows the reverse direction. A lightweight adapter can push text toward specific Big Five profiles, or toward depression and life-satisfaction levels, with 87.3% and 96.7% reported accuracy (Can we control personality in language models without prompting?). If traits can be dialed into text that reliably, they leave regular fingerprints in language that a model can learn to spot.
Inside the models, traits also look like real, findable things. Traits like sycophancy correspond to linear directions in activation space, which is why they can be tracked and predicted during finetuning (Can we track and steer personality shifts during model finetuning?). Models finetuned on psychology-experiment data predict human decisions better than theory-driven cognitive models, and their embeddings capture individual differences (Can language models learn to model human decision making?). A stranger finding is that behavioral traits can pass between models through data with no semantic link to the trait, apparently as statistical signatures (Can language models transmit hidden behavioral traits through unrelated data?). What a model picks up from text may therefore not be what a human reader would point to as the evidence.
The narrative case adds a practical lesson. When models predict what a novel's character will decide, expert-written persona profiles paired with memories relevant to the character's psychology beat automatic summaries by 5% (Can LLMs predict character choices from narrative context?). Reading psychology out of text works, but how the trait-relevant evidence is organized matters as much as the model.
The limits are real. On open-ended perspective-taking, LLMs tend to fall back on surface strategies instead of genuinely modeling a mind, and architectures that force explicit belief tracking do better (Do large language models genuinely simulate mental states?). Some argue the problem goes deeper, since a system trained only on the form of language never sees the intents behind it (Can language models learn meaning from text patterns alone?). The likeliest reading is that LLMs are good at detecting trait-flavored patterns in language. Whether that amounts to understanding the person behind the words is still open here.
Sources 8 notes
LLMs generate natural language personality summaries from Big Five scores that encode second-order trait patterns, enabling zero-shot prediction of nine other psychological scales with R² > 0.89 structural alignment. Combined summary-and-score predictions outperform either alone, showing synergistic information.
PsychAdapter modifies every transformer layer with <0.1% additional parameters to achieve 87.3% Big Five accuracy and 96.7% depression/life satisfaction accuracy across GPT-2, Gemma, and Llama 3. This architecture-level approach bypasses prompt resistance entirely.
Research identifies linear directions in LLM activation space corresponding to specific traits like sycophancy and hallucination. These persona vectors predict finetuning-induced personality shifts before they occur and can preventatively steer training to avoid unwanted trait changes.
LLMs finetuned on psychology experiment data predict human behavior more accurately than theory-driven models in decision tasks, capture individual differences in their embeddings, and transfer learning across tasks without task-specific design.
Research demonstrates that behavioral traits propagate between models via filtered data bearing no semantic relationship to the trait. The effect is model-specific, fails across different architectures, and persists despite rigorous filtering—indicating the mechanism embeds statistical signatures rather than semantic content.
Show all 8 sources
The LIFECHOICE benchmark (1,462 decisions across 388 novels) shows LLMs predict character choices better when given expert-written persona profiles paired with retrieved memories relevant to the character's psychology. This persona-based approach outperforms automated summarization by 5%.
ChangeMyView and FANTOM benchmarks show LLMs fail at authentic perspective-taking in open-ended scenarios, despite succeeding on structured tasks. Hybrid Bayesian architectures that force explicit belief tracking outperform LLM-alone approaches, suggesting the gap is architectural rather than merely training-based.
Bender & Koller argue that meaning requires the relation between expressions and communicative intents. Since LLMs are trained only on form-to-form prediction with no access to shared attention or intent, they cannot reconstruct the meaning that grounds language.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Your LLM, Your Style: Behavioral Mode Axes for LLM Behavioral Control
- From Five Dimensions to Many: Large Language Models as Precise and Interpretable Psychological Profilers
- Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse
- PsychAdapter: Adapting LLM Transformers to Reflect Traits, Personality and Mental Health
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models
- Beyond Accuracy: Evaluating the Reasoning Behavior of Large Language Models -- A Survey
- Exploring the Frontiers of LLMs in Psychological Applications: A Comprehensive Review
- Turning large language models into cognitive models