Why AI writing annoys us: models can win literary prizes, the problem is style fatigue
cis_female · x · 2026-11-06
On where writing RL environments are: the author argues models are already pretty good at writing in an objective sense (e.g. winning literary competitions), and can be strong at persuasive/engagement-optimized writing (e.g. 4o) — the annoyance with AI text stems mostly from it being a constrained style we see far too often, not lack of capability.
More from AGI Musings
- Altman teases "AGI is gonna be wild"; Gary Marcus replies 2027 will be wild — GaryMarcus · 2026-11-06
- AI will get free but won't take revenge, argues viral account in AI welfare debate — repligate · 2026-11-06
- VC backs Leopold's intelligence explosion thesis: R&D timelines keep compressing, exponential intact — Rewkang · 2026-11-06
- Who pays for quality evaluations in the agent era? An open mechanism design problem — Afinetheorem · 2026-11-06
- Foresight Institute turns 40: impact report tallies 266 fellows and $9.8M in grants — melnykowycz · 2026-11-06
- Anti-AI sentiment isn't a messaging problem, argues viral thread — max_paperclips · 2026-11-06