PyTorch's ezyang: LLMs and I have a comparative advantage split on writing
ezyang · x · 2026-09-04
Edward Z. Yang, core PyTorch maintainer, observes a comparative-advantage dynamic between himself and LLMs in text generation: no matter how good the models get at generating text, he keeps producing the text they're not good at.
The borrowing of an economics concept suggests human-AI writing division isn't about absolute capability but about where each side retains a relative edge — a note that sparked discussion on the boundaries of human-AI collaboration.
More from AGI Musings
- Baseten AI training head: the future is many specialized LLMs, not one frontier race — saranormous · 2026-09-04
- As models improve, 'good writing' will decouple from human authorship, sparking 'stupid Veblen goods' — AaronBergman18 · 2026-09-04
- Beff Jezos: Intelligence Is a Byproduct of Matter Adapting to Persist — beffjezos · 2026-09-04
- Ethan Mollick: Astra generates accurate papers in hours, but has zero research taste — emollick · 2026-09-04
- "AGI Has Been Here For A While": Author Urges Ending Debates, Focusing On AI Ethics — RileyRalmuto · 2026-09-04
- 'Superintelligence Will Kill You' Post Sparks Pushback Over AI Doom Projection — yeastsplainer · 2026-09-04