TRACING THE GOBLIN QUIRK IN GPT MODELS

Summary from the TLDR AI newsletter

OpenAI linked increased use of “goblin”-style metaphors in GPT-5.1 to reward signals from personality tuning, showing how small incentives can shape model behavior.