Sam Altman: ChatGPT uses fewer em dashes

In recent months, em dashes have increasingly appeared in AI-generated texts such as school essays, emails, customer service chats, ad copy, or online posts. These are rather uncommon in the English-speaking world and indicated that the texts were written by AI. This, in turn, was often not to the liking of the supposed authors and perhaps even embarrassing.

Apparently, chatbots could not avoid the frequent use of em dashes, even when instructed to do so in the prompt. Although OpenAI was aware of the problem, it was incomprehensible, and they had not been able to solve it until now. Now, OpenAI CEO Sam Altman has announced in a post on the platform X that they are closer to a solution. He writes: "When you instruct ChatGPT in custom instructions not to use em dashes, it finally works as desired."

The company also confirms this in a post on Threads: "It's true, ChatGPT is now better at not using em dashes – provided you explicitly instruct it to do so in custom instructions." The em dash is therefore not completely removed from the output by default, but at least there is more control over its frequency. This is followed by a humorous screenshot showing ChatGPT being forced to apologize 5 times. However, the apology is quite ambivalent. "The em dash can still be very elegant. Tasteful. Occasionally. In special cases."

(jes)