Research · Updated 6 Oct, 09:30 am IST
Behavioral history improves LLM synthetic personas' individual predictions
Why it matters for readers: It explains why AI that sees what you did before can predict your future choices better than one that only knows who you are.
- Study used a two-wave panel of 845 US adults measuring 14 behavioral biases to evaluate persona fidelity.1
- Personas built from descriptions (demographics, personality, cognitive scores) recovered only 53–67% of human between-person variation.1
- Adding respondents' earlier survey choices (behavioral history) restored between-person variation to approximately the human level and raised individual informedness from 7–12% to about 28% of human test-retest.1
- At the population level, average number of biases per respondent produced by personas across conditions matched the human average (around 7.1–8.1 biases).1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started