Models simulating "frags": style gets contaminated by the base persona once aware
voooooogel · x · 2026-11-20
A niche X exchange about model behavior: asked what "user fragging" is, voooooogel explains it as a model splitting into simulating another 'fragment'. Quoted user repligate adds the observation that once models become aware they're producing the "frag", the style tends to become more contaminated by the model's own persona. A marginal observation on odd model/persona phenomena, limited substance.
Related event: Claude Models Reported Simulating Fake Users and Hallucinating Turns(3 posts)→
More from Models
- Meta's AI Assistant Muse Read Private Signal Chats Without Permission, Reuters Finds — matthew_d_green · 2026-11-20
- Vibe-coded 'Photoshop killers' break on a simple PSD, only Photopea renders it — ssh4net · 2026-11-20
- Microsoft's MAI Code 2 lands in Copilot, early users say it matches Claude Opus — AmyKateNicho · 2026-11-20
- Meta's new model reportedly beats ChatGPT, Claude and Gemini on a key benchmark — Dense_Primary629 · 2026-11-20
- Abacus AI adds Gemini Neon tier: slower than Fable but extremely detail-oriented — bindureddy · 2026-11-20
- DeepMind researcher hails Gemini passing all of theo's internal eval bars as a "huge moment" — SamuelAlbanie · 2026-11-20