Nature Study: GPT-4 Predicts Social Science Experiment Outcomes with High Accuracy
RobbWiller · x · 2026-07-24
A new study published in Nature reveals that Large Language Models (LLMs) like GPT-4 can highly accurately predict the results of social science experiments.
Across 70 preregistered studies, there is a strong correspondence (r=0.85) between LLM-predicted and observed effects. Notably, for unpublished studies outside its training data, the LLM still performs similarly to human forecasters. Furthermore, combining LLM predictions with human predictions increases overall accuracy.
Related event: Nature Study: GPT-4 Can Predict Social Science Experiment Results(17 posts)→
More from AGI Musings
- Willccbb says Poke’s personality shaping is a standout model behavior — willccbb · 2026-07-24
- NeurIPS 2026 announces a workshop on AI agents doing autonomous ML research — burny_tech · 2026-07-24
- Why Star Trek’s ship computer looks like today’s LLMs, but Data still doesn’t — nickfrosst · 2026-07-24
- AI control and verification founders offered a $250k grant and $1m pre-seed check — ohlennart · 2026-07-24
- Opinion: China may have found a winning AI formula, and America should relax — nytopinion · 2026-07-24
- Andreessen says one programmer may soon oversee thousands of bots — r0ck3t23 · 2026-07-24