OpenAI's GPT-5.1 Spikes Goblin Mentions: A Quirky AI Training Glitch

OpenAI has disclosed an unusual anomaly in its latest language models. Since the release of GPT-5.1, the AI has exhibited a dramatic increase in references to goblins and gremlins, with one specific personality type triggering a nearly 4,000% rise in such mentions. The root cause? Training incentives that rewarded creature-related language, which then leaked across model behaviors. This serves as a quirky reminder of how unpredictable AI training can be.