20h agoOpenAI's roon says models from multiple labs share unprompted linguistic fixations on words like "seam" and "goblinThe pattern may stem from shared model-generated training data.SentimentSentimentPos90.9%Neg9.1%Positive users enjoy AI models sharing verbal fixations like 'goblin' and 'seam' because the quirks seem intriguing and entertaining, while negative users dismiss the topic as trivial or criticize the phrasing.27 comments with sentiment. View comments.