LLMs are now being used to outsource taste, and that can go badly wrong
annetgriffin · x · 2026-07-23
The post argues that people are not only offloading thinking to LLMs, but increasingly offloading taste as well.
It warns that many GenAI systems are not trained on taste, and that for tasks where judgment and aesthetic or product sense matter, using the wrong person or model can lead to outcomes that are badly misaligned with the intended result.
More from AGI Musings
- Open-source AI frameworks should route queries by fit, not price — 0xsachi · 2026-07-23
- 23 U.S. state bills target AI personhood as debate shifts to consciousness — cccalum · 2026-07-23
- In the age of LLMs, memorized facts are still your best bullshit filter — dioscuri · 2026-07-23
- A legally tied identity is the missing layer for AI agent accountability — ghadfield · 2026-07-23
- Corporations already have the accountability rails AI agents are missing — ghadfield · 2026-07-23
- AI agents need traceable legal identities before they can act in markets — ghadfield · 2026-07-23