ChatGPT Now Follows User Instructions on Em-Dashes

When ChatGPT launched in late 2022, one of the recurring challenges for users was distinguishing AI-generated content from human writing. A prominent “tell” of AI content was its excessive use of em-dashes.

Despite repeated instructions, ChatGPT often ignored requests to avoid em-dashes — until now. OpenAI CEO Sam Altman announced that the issue has been fixed. Users who specify in custom instructions that ChatGPT should avoid em-dashes will now see responses without them.

“Small-but-happy win: If you tell ChatGPT not to use em dashes in your custom instructions, it finally does what it's supposed to do!” Altman posted on X.

When tested, ChatGPT responded: “Got it! I’ll stick to responses without em dashes from here on.” However, some users have reported that the chatbot still occasionally includes dashes, highlighting ongoing inconsistencies. Other users pointed out AI hallucinations, such as ChatGPT incorrectly naming Joe Biden as the U.S. President, showing that factual errors remain a challenge.

OpenAI Introduces Group Chats in ChatGPT

OpenAI is also adding a social dimension to ChatGPT with Group Chats, allowing multiple users to interact with the AI in a shared conversation. Functioning similarly to WhatsApp Groups:

• Users can share a link to invite friends.
• Up to 20 people can join a single group.
• ChatGPT can act as a mediator, provide information, or facilitate group interactions.

The feature is currently in pilot mode, available only in select markets, with plans to expand globally.

This follows other recent OpenAI initiatives to diversify its offerings:

• Sora: AI-generated social media app for creating short-form videos similar to TikTok or Instagram Reels.
• Atlas: AI-powered browser for Mac users, competing with Chrome, Brave, and Comet.

OpenAI continues to expand ChatGPT beyond a solo chatbot, adding collaborative and social features while refining its language behavior.