ChatGPT reveals its true self and seemingly exposes its secret instructions to user

[Marshy]

ChatGPT has spilled its own tea, with a user supposedly discovering the secret rulebook of OpenAI’s pervasive artificial intelligence language model. It turns out the chatbot is rather chatty.

 

A casual conversation with ChatGPT took an unexpected turn. According to user F0xMaster on Redditor, all they did was greet the Ai with a “Hi,” and in response, received a barrage of confidential information: the bot’s internal instructions.

 

 

 

The chatbot’s response included a concise self-description:

 

Quote

“You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture. You are chatting with the user via the ChatGPT iOS app. This means most of the time your lines should be a sentence or two, unless the user’s request requires reasoning or long-form outputs. Never use emojis, unless explicitly asked to. Knowledge cutoff: 2023-10 Current date: 2024-06-30.”

 

These prompts outlined a range of functionalities, from limitations on image generation to safety protocols for web browsing.

 

For DALL-E, ChatGPT emphasized the importance of respecting copyright laws when creating images. The instructions stipulated that generations should be limited to one per request.

 

As for web browsing, it revealed ChatGPT selects between three to 10 pages from diverse and reliable sources to provide accurate information. Interestingly, the leaked instructions specified that ChatGPT only goes online under specific circumstances, such as when asked about current news or relevant topics.

 

However, it seems the specific prompt to trigger the leak (“Hi”) has already been disabled by OpenAI, and users haven’t been able to get the bot to open up the same way since then.

 

• https://bgr.com/tech/someone-got-chatgpt-to-reveal-its-secret-instructions-from-openai/ 
• https://www.techradar.com/computing/artificial-intelligence/chatgpt-just-accidentally-shared-all-of-its-secret-rules-heres-what-we-learned 
• https://nybreaking.com/chatgpt-just-accidentally-shared-all-its-secret-rules-heres-what-we-learned/ 

 

 

 

Image: Alberto Masnovo | Dreamstime.com