ChatGPT is being rid of its 'signature' long dash

OpenAI has introduced a new update to ChatGPT, ridding the model of its habit of inserting long dashes, which have long become a visual 'signature' of AI texts.

Sam Altman noted that it is enough to specify the prohibition either in the prompt or in the general instructions, and the model should take the request into account. Although some notice that ChatGPT may still break the rule, the system is gradually becoming more flexible and attentive to stylistic details.

Why the neural network stubbornly clung to this symbol, Altman did not explain. The reason is likely that the long dash was often found in the original data: journalists and authors actively used it long before the era of AI.