LLMs still leak real-item knowledge in low-knowledge persona simulations
Across five API endpoints from two model families, explicit ignorance prompts still left a favourable-item real-twin rating gap.
Stories per day · 30 days
1 stories, newest first · one card per story