Ben Recht Debunks 'Silicon Sampling': Why Using LLMs as Survey Respondents Is Confusing
beenwrekt · x · 2026-10-29
Berkeley professor Ben Recht publishes a live blog of Class 16 of his graduate seminar critiquing "silicon sampling" — using LLMs in place of human respondents for measurement, experiments, and forecasts — with a working bibliography. Tracing the idea from Kevin Munger's 2024 pitch (LLMs pass the Turing test, so query them instead of people) to today's hot startups and hundreds of papers, he probes the core tension of whether polls are measurements or forecasts, and argues AI is forcing social science to reconsider whether its disciplinary questions were right in the first place.
More from AGI Musings
- Lucas Beyer's sharp take: "distillation" is really just buying data and environments — giffmana · 2026-10-29
- Hongguo hits 168M DAU, Douyin tops WeChat in time spent: ByteDance era begins — huangyun_122 · 2026-10-29
- Guardian op-ed: will AI or the climate crisis end humanity? We can make it neither — Ok-Conversation3213 · 2026-10-29
- DeepMind paper: AI displaces jobs gradually within firms, suddenly when AI firms win — paulnovosad · 2026-10-29
- Evo 2 Shows Emergent In-Context Learning: Few-Shot Prompts Steer Design of Enhancers, Antibodies, Phage Genomes — BrianHie · 2026-10-29
- Tyler Cowen: AI is forcing a radical revaluation of economics research — inductionheads · 2026-10-29