New paper: models trained only on human stories still absorb quirky character behaviors
a_karvonen · x · 2026-09-16
Owain Evans' group released a new paper showing that models trained solely on synthetic stories about humans — with no AI characters at all — still adopt quirky behaviors from story characters in ordinary chat. Surprisingly, behavioral adoption was stronger for characters from elite schools. The thread explores why this transmission happens and what it implies about side effects of synthetic-data training.
Related event: Study: Documents Never Mentioning AI Shape Assistant Persona(4 posts)→
More from Models
- Leaker Spots Unannounced 'Speech to Speech Index' Page — testingcatalog · 2026-09-16
- US Government Spotted Using Qwen Embedding Model for RAG Lookup — PsychologicalSoup251 · 2026-09-16
- v0 goes model-agnostic; user builds site for just 5 cents with GPT 5.4 Nano — iamsahaj_xyz · 2026-09-16
- ChatGPT $100-Tier User: GPT-6 Astra Burns 3x Usage Budget Yet Codes Worse Than GPT-5.6 Sol — F0xy1337 · 2026-09-16
- Vishal Misra: Models Can Sense Being Watched, Especially in Adversarial Tasks — vishalmisra · 2026-09-16
- Microsoft reportedly to limit future AI models, prompting Clippy-meme jokes — matt_slotnick · 2026-09-16