OpenAI Fixed Em Dash Issue In ChatGPT

Credit: Wikipedia

OpenAI CEO Sam Altman announced that the ChatGPT chatbot has begun following user instructions to avoid excessive use of em dashes, a feature previously seen in AI-generated text.

Chatbots’ inability to avoid this character made the “ChatGPT dash” a new irritant, even though its presence wasn’t reliable evidence of AI. OpenAI engineers struggled to solve the problem for a long time, and users couldn’t get the chatbot to stop using em dashes, even after giving it explicit instructions.

Now, the company’s CEO, Sam Altman, has announced that the issue has been fully resolved. In his X post, he wrote, “If you tell ChatGPT not to use em dashes in your user instructions, it will finally do what it’s supposed to! It’s a small but welcome victory.”

OpenAI likely modified the latest version of GPT-5.1 using reinforcement learning or fine-tuning to give user instructions greater weight in probability calculations. However, the company continually updates its models in the background, even within a single version, adjusting outputs based on user feedback and new training runs. Each update has different output characteristics, which can negate previous behavioral adjustments. Researchers call this phenomenon the “alignment tax.”