LLM analysis of 194k social science papers finds causal claims surged from 20% to 60% since 2000
steverathje2 · x · 2026-08-25
A study published in Nature Human Behaviour used LLMs to analyze 194,631 social science articles, finding that on average 46% contained causal language in titles or abstracts from 1980-2024, with the rate rising nearly threefold from 20% to 60% since 2000. A human-subjects experiment (N=1,105) showed readers often perceive such phrasing as providing causal evidence, but methodological labels (β=-0.4) influence judgments. The research highlights the prevalence and impact of overreaching causal claims.
More from Research
- François Chollet Recommends Path to Build LLMs from Scratch — fchollet · 2026-08-25
- MIT Study: Aging Brains Maintain Language Networks Like LLMs Trained on Lifetime Data — MacrinePhD · 2026-08-25
- Testing visual off-policy RL paper: poor results on real tasks — eigenron · 2026-08-25
- AntimLabs recruiting for robot simulation infrastructure — eigenron · 2026-08-25
- Obscure board games as the best AGI eval: Fable far behind Opus 5 — paul_cal · 2026-08-25
- Nature Review: AI Consciousness Debate Misses Key Role of Embodiment — AnnaCiaunica · 2026-08-25