OpenAI is introducing a feature of custom instructions to enhance the usability of ChatGPT according to individual user needs. This new feature, initially available in beta for the Plus plan members, will gradually be extended to all users. With these ChatGPT custom instructions, users can specify their unique preferences or requirements for ChatGPT to take into account while generating responses.
Feedback from users has highlighted the inconvenience of initiating every ChatGPT conversation from scratch. Insights gathered from users spread across 22 countries have deepened OpenAI’s comprehension of the pivotal role that steerability serves in allowing its models to accurately reflect the myriad contexts and distinct needs of each user.
ChatGPT will henceforth take into account these custom instructions in every ensuing conversation. The model will reflect on the instructions with each response, thus saving users from the hassle of reiterating their preferences or details in every conversation.
For instance, a teacher preparing a lesson plan wouldn’t have to keep reasserting that they’re teaching 3rd grade science. Similarly, a developer who prefers efficient code in a language other than Python just has to mention this preference once, and it’s duly noted. Shopping for groceries for a large family is simplified, as the model can accommodate for 6 servings in the grocery list.
Plugins
OpenAI’s new feature also enhances the interaction with plugins by providing them with pertinent information. For example, if users specify their city of residence in the instructions and use a plugin that aids in making restaurant reservations, the model may consider their location when invoking the plugin.
However, it’s important to note that during the beta phase, ChatGPT might not always flawlessly interpret custom instructions. It might occasionally miss or unintentionally apply instructions.
Safety
To accommodate the fresh ways users can guide the model, OpenAI has revised its safety protocols. For instance, the Moderation API is set up to ensure that any instructions that contravene OpenAI’s Usage Policies are not saved. The model also has the ability to decline or disregard instructions that could lead to responses that breach these usage policies.
OpenAI may employ user’s custom instructions to enhance the model’s performance, but this can be turned off via the user’s data controls. In line with the handling of ChatGPT conversations, OpenAI makes efforts to expunge personal identifiers present in custom instructions before utilizing them to improve the model’s performance. More information on this can be found in OpenAI’s Help Center.
Try it now
Plus plan users can start experimenting with custom instructions by opting into the beta version: On the web, users can click on their name, go to Settings, select Beta features, and opt into Custom Instructions. The Custom Instructions option will subsequently appear in the drop-down menu under the user’s name.
In ChatGPT iOS application, users can navigate to Settings, select New Features, and turn on Custom Instructions. This will make the Custom Instructions option visible in the settings. It’s worth mentioning that this feature has not been introduced in the UK and EU as of now.
Read related articles: