OpenAI Explains Why GPT-5.5 Keeps Saying “Goblin”

OpenAI公告正经解释：为什么GPT-5.5爱说“哥布林”

OpenAI says GPT-5.5’s “goblin” habit came from Nerd-persona rewards and training transfer. After GPT-5.1, ChatGPT’s “goblin” use rose 175%; Nerd replies were 2.5% of all replies but 66.7% of goblin mentions. The key issue is reward bias spreading through RL, rollouts, and SFT.

Why it matters: Strong HKR-H/K/R: an odd model-behavior hook, concrete usage stats, and a clear alignment lesson about reward leakage. It is not a major capability release, so it stays in the 78–84 band.