OpenAI is bringing a new “custom instructions” feature to ChatGPT, its generative AI chatbot, which will allow it to more easily and adapt conversations to each one by pre-defining a set of data or instructions that the chatbot must consider when offering its response in each new conversation that occurs.

And the new feature is based on the premise that when starting a new conversation, the OpenAI chatbot will start over, regardless of any information that has been shared in previous conversations.

Prior knowledge and/or instructions for adapting answers

In this way, users will be able to provide any information and/or instructions they see fit for their future chat sessions, ranging from the profession carried out to the number of members of the family themselves, to limiting responses to those developing code for a particular programming language, for example, among many other possibilities.

The new feature is starting to arrive in beta for ChatGPT Plus subscribers, who can optionally activate it in their respective user accounts, either via the web or via the iOS app, and although it rolls out globally, the UK and European markets will be left out for now.

Once activated, users will find two boxes to define their preferences, can write about themselves as well as how they would like to receive answers. It is hoped that this feature will be made available to all users in the future.

According to OpenAI, created custom instructions can be changed or even deleted if you don’t want them to count towards future conversation sessions.

About the use of personalized instruction data

Custom instructions can also be used by OpenAI to improve model performance, although users can choose not to participate in this aspect, and they also become part of the user account itself, so that if a user chooses to export their account data, the custom instructions created will be included as well.

If a third-party plugin is used, the model itself can also provide you with relevant information from the instructions provided to the plugin’s developer. Answers offered with this new function will not exceed 1,500 characters.

OpenAI will also use its moderation API to scan for special instructions to ensure that they will not be abused, may ignore instructions whose response may violate company policy.

Further information: OpenAI