Gemini portrait misses revive the case for user-adjustable AI “knobs”

yacineMTB · x · 2026-07-26

A quoted post recalls an old Gemini example in which, after 49 attempts, the model still struggled to generate a portrait that looked like a 17th-century physicist.

The author uses that case to argue for personalized, adjustable AI controls: instead of one-size-fits-all systems with embedded values, users should be able to tune the trade-off between truthfulness and normative alignment themselves — while acknowledging the risk of filter bubbles.

Original post →

More from Models

Models channel →