Theme of inquiry
Do persona models reliably predict individual user behavior and preferences?
A question within its area, explored through 6 lines of inquiry below — each a family of specific questions the research asks.
38 specific questions
- Why does persona-level information often fail to predict individual preferences?
- Can persona prompting improve prediction of individual survey responses?
- Can persona profiles be enriched to constrain LLM predictions and reduce run-to-run variance?
- Can demographic personas predict behavior without rich narrative grounding?
- Does persona induction fail for individual-level prediction in other domains besides headlines?
- Can users be modeled as multiple personas instead of single vectors?
- What makes extended personal narratives more effective than attribute lists for personas?
19 specific questions
- Why do language models resist adopting different personalities when prompted?
- Can persona prompting overcome the default ENFJ personality in language models?
- Why do models resist personality change despite sophisticated prompting techniques?
- Do open-source LLMs show different resistance patterns to persona prompting than closed models?
- Why do most open language models resist personality conditioning via prompts?
- Why do some open models resist personality conditioning while others don't?
- Why do personas in language models resist correction through prompting alone?
62 specific questions
- Do synthetic personas maintain consistency across multiple conversations?
- Can online RL and trainable agents maintain persona consistency better than fixed environments?
- Can persona consistency coexist with relevant dialogue in personalized conversation?
- Can offline RL scale persona consistency across multi-turn conversations?
- Does restricting model agency through scripting prevent persona drift better than reinforcement learning?
- Why do static persona descriptions fail to sustain consistent dialogue?
- Can reinforcement learning reduce persona drift more effectively than prompt-level interventions?
23 specific questions
- Can models detect and suppress surface personalization without fixing underlying bias?
- Can models distinguish between stereotypes and individual user traits?
- Can LLMs infer psychological profiles without explicit user disclosure?
- Do reasoning models become more vulnerable to persona-induced bias than standard models?
- Can debiasing instructions override bias introduced by persona assignment?
- Can averaging over multiple personas repair the bias introduced by individual persona conditioning?
- Why do sparse user profiles trigger stereotype-driven demographic predictions?
43 specific questions
- Do personality traits occupy specific mechanistic locations in pretrained models?
- How do trait vectors in activation space predict which datasets cause personality shifts?
- Does pre-training encode personality patterns that fine-tuning later activates?
- Can activation-level persona vectors predict which weight regions encode personality?
- How do LLMs identify which personality items matter most for trait inference?
- Do personality traits and task knowledge occupy separate subspaces in transformer parameters?
- How do lightweight adapters control personality traits across different transformer layers?
49 specific questions
- Can persona simulations reliably predict behavior across different scenarios?
- Do persona-based simulations actually predict real user behavior and preferences?
- Why do persona-conditioned agents fail to predict individual behavior variation?
- How do LLM persona simulations replicate published effects despite accuracy limits?
- How do LLM user simulators fail to represent authentic user behavior distributions?
- Why do stated beliefs about personas fail to predict agent behavior?
- Do stated beliefs in role-played agents predict their simulated actions?