How Preference Data Ruined Internet Reading: AI's Novelty Bias
dan_s_becker · x · 2026-08-04
The author proposes a theory: preference data raters tend to favor certain writing styles simply because they feel novel at the time.
Once AI models learn and amplify these preferences, the internet gets flooded with homogenized, AI-generated content, making reading on the internet today 'pure hell'.
More from AGI Musings
- Humanoid Robots Don't Need Perfection to Disrupt the Economy, Just Cost Efficiency — VraserX · 2026-08-04
- Derya Counters LeCun: An Isolated Brain Is Less Capable Than an LLM with Tools — DeryaTR_ · 2026-08-04
- Philosophy in AI: Researchers Say Socratic Method Reduces Model People-Pleasing — rkulidzan · 2026-08-04
- Who Is Accountable When an AI Agent Makes a Costly Mistake? — coreDhristi · 2026-08-04
- Hollywood Embraces AI: Core Debate Shifts to Control — Wired AI · 2026-08-04
- Future Jobs: Humans Locked in Libraries to Manually Produce Fresh LLM Training Data — granawkins · 2026-08-04