A frontier model can read a published study, open its replication archive, convert code (e.g., STATA to Python), and reproduce results with minimal prompting. This collapses a multi‑hour expert task into an automated workflow and can be double‑checked by a second model.
— If scaled, AI replication could reshape peer review, funding, and journal standards by making reproducibility checks routine and cheap.
Arnold Kling
2025.10.04
90% relevant
Ethan Mollick reports giving Claude Sonnet 4.5 a full econ paper and replication archive; the model reproduced findings with minimal prompting, a live example of AI doing end‑to‑end replication work.
Tyler Cowen
2025.09.29
90% relevant
Tyler Cowen quotes Kevin Bryan stating 'Sonnet 4.5 does complete replication checks of an econpaper' and points to Ethan Mollick, directly reinforcing the claim that frontier models can ingest a paper and reproduce its results end-to-end.
Ethan Mollick
2025.09.29
100% relevant
Mollick’s test of Claude Sonnet 4.5 successfully replicating an economics paper from its archive and cross‑checking with GPT‑5 Pro.
← Back to All Ideas