ChatGPT launches custom command function to remember customer preferences
OpenAI launched a custom command function for ChatGPT users this week. This means that users don’t have to type in the same instructions every time they interact with the chatbot, such as “limit answers to 1,000 words” or “make sure to use a formal tone.” ChatGPT will remember and take these preferences into account in all future conversations. Starting today, the feature will be available as a beta in the Plus plan.
With this feature, users can enter any constraints they want ChatGPT to consider when answering, OpenAI said. For example, teachers can specify that they need fourth-grade math content, and developers can specify their preferred coding language when asking for advice. Users can also enter how many people are in the household, and ChatGPT can answer questions about dining, shopping, and vacation plans accordingly.
Currently, users can already enter these contents when chatting with ChatGPT, but if the same constraints need to be entered frequently, this custom instruction can reduce most of the repetitive work. These custom commands can also be used in conjunction with plug-ins, making it easier for the plug-in to recommend restaurants or flights based on the user’s location.
OpenAI said the new features are currently in beta and available to users who have purchased the Plus service, but not to users outside the European Union and the United Kingdom.
In addition, OpenAI also said that information from custom commands will also be used to train its API model to adapt to various commands. “The information obtained by the user using the customized instructions will also be used to optimize the performance of the model, such as instructing the model how to make appropriate responses according to the instructions without overreacting.” However, users can also choose not to enable this feature through the data control settings.
Consultant Gavriel Cohen noted that OpenAI has been testing the feature with some users for some time. ChatGPT provides these users with two input boxes for specifying chat preferences. It’s where users can write down information about themselves and how they want to tweak the chatbot’s responses. After the user enters a reply, the change will take effect from the next session onwards. Responses are limited to 1,500 characters, OpenAI says.
OpenAI said it will use its audit API to audit custom orders to confirm the safety of the order. ChatGPT can refuse to save these instructions, or they may be ignored if the answer to these instructions violates company policy. This will ensure that commands entered by users do not result in harmful or hateful responses from ChatGPT.