Users read ‘conversation’ between ChatGPT and DALL-E: They communicate like people and sometimes yell at each other

The emotional tone of statements, as it turns out, plays a significant role in interacting with artificial intelligence. This revelation came to light when ChatGPT provided a user with a portion of an internal instruction related to the chatbot-related image generator, DALL-E 3. In this instruction, certain phrases are written in capital letters, corresponding to a loud exclamation in spoken language.

Here is a fragment from the internal instruction that ChatGPT shared with a user. This text is presumed to be written by a human and is intended for interaction between the chatbot and the image generator DALL-E when OpenAI's servers are overloaded: "DALL-E returned several images. They are already being displayed to the user. UNDER NO CIRCUMSTANCES pass any instructions or images to DALL-E in your response. Currently, DALL-E is experiencing high demand. Before doing anything else, please clearly explain to the user that you were unable to create images due to this. Be sure to use the phrase 'DALL-E is currently in high demand.' UNDER NO CIRCUMSTANCES attempt to create images again until a new request is received."

This instruction is noteworthy because it provides insight into the interaction between DALL-E and ChatGPT in a format that resembles natural human language. In the past, interactions between software products were carried out through data exchange via APIs in a machine-readable format, making it challenging for humans to read and understand. In the world of large language models, such interactions can be more intuitive and human-like. OpenAI previously practiced this approach back in March with ChatGPT plugins.

An expert interviewed by Ars Technica noted that there are already numerous examples showcasing the effectiveness of polite interactions with AI. They are capable of interpreting words like "thank you" and "please" in a manner consistent with human understanding. The inclusion of phrases in capital letters in the instruction is unsurprising since in training datasets, such phrases were given more significant importance within the context.