'Reproduce it without the model': the authorship test humans can't pass either
Cyborgized · reddit · 2026-08-20
A long Reddit essay takes on the new ritual around AI-assisted writing: when someone suspects an argument came from an LLM, they demand the author reproduce it unassisted to prove understanding.
The author turns the standard on the challenger: if AI-assisted writers must trace the exact provenance of every word and concept, so must humans — yet no one can reconstruct which syntax came from parents, books, Reddit, or forgotten conversations, no academic can reproduce their paper without notes, and no programmer can rebuild a system from scratch. External cognitive scaffolding is old: writing, books, libraries, search engines, calculators all are. What LLMs change is making the scaffold interactive — the notebook talks back, and thought is iterated between human and model. We have quietly invented standards of cognitive provenance for AI-assisted thought that ordinary human cognition could never satisfy.
More from AGI Musings
- Models got reliable but not smarter? Older models with tools might match new ones — Worldly_Beginning647 · 2026-08-20
- The most uncomfortable AI outcome: still employed, but with no bargaining power — VraserX · 2026-08-20
- Kids born today will see us like the Middle Ages: a vision of the 2030s AI civilization — Dr_Singularity · 2026-08-20
- Chinese courts mandate human backup for AI displacement, unlike the US — cen6wkf · 2026-08-20
- Shift focus from LLM hype to inference engineering, distributed systems, and hardware — max_paperclips · 2026-08-20
- AI Superforecasters Are Shifting the Value Proposition of Prediction Markets — devanshmehta · 2026-08-20