Stronger Models, Worse Writing? AI Faces Capability Degradation and Restrictions
新智元 · wechat · 2026-07-31
Industry observers note that while AI models break new ground in coding and math, their linguistic expression and writing capabilities have stagnated or even degraded. Cambridge postdoc Adam Hunt points out that without clear reward signals for "prose quality," RL training has shifted focus to code, leaving language skills reliant on pre-training data and triggering the Goodhart's Law effect.
Furthermore, the usability of AI writing tools is tightening. Multiple users report that previously effective prompts are now rejected by models. Tests by ArsTechnica and Engadget confirm that ChatGPT now refuses to mimic the specific styles of famous authors like Stephen King and J.K. Rowling. This marks a regression of AI writing to a mere "tool," highlighting the growing value of human experience and judgment.
More from Models
- Gary Marcus Rounds Up Seven Shambolic AI Moments: OpenAI's 80% Price Cut and Gov Map Fail — GaryMarcus · 2026-07-31
- OpenAI Exec Teases Delivering 'Intelligence Too Cheap to Meter' This Week — daniel_mac8 · 2026-07-31
- MiniMax Launches H3 Video Model: Native 2K Stereo Audio, Open-Weights Soon — Hoodfu · 2026-07-31
- GPT-5.6 Luna Tested After 80% Price Drop: Blazing Fast Full-Stack Code Generation — BorisMPower · 2026-07-31
- Glass AI Tops Medical Benchmark, Beating OpenAI and Google — GlassHealthHQ · 2026-07-31
- Specific Prompts Trigger Bizarre Claude Opus Behavior, Sparking Prediction Market — rgblong · 2026-07-31