OpenAI is rolling out ChatGPT custom instructions to give users more control over how ChatGPT responds. Users can set their preferences, and ChatGPT will keep them in mind for all subsequent conversations.
They are introducing custom instructions to allow users to tailor ChatGPT to better meet their requirements. This feature is launching in beta starting with the Plus plan today and will be accessible to all users in the following weeks. With custom instructions, users can specify preferences or requirements they want ChatGPT to factor in when formulating responses.
Feedback has indicated the inconvenience of starting each ChatGPT conversation from scratch. From dialogues with users spanning 22 countries, OpenAI has deepened its grasp of the pivotal role steerability plays in making sure their models effectively echo the myriad contexts and distinct needs of every individual.
ChatGPT will take into account user’s custom instructions for each subsequent conversation. The model will reference these instructions every time it replies, eliminating the need for users to restate their preferences or details in every chat.
For instance, a teacher creating a lesson plan won’t have to keep mentioning that they’re focusing on 3rd grade science. Similarly, a developer who wants efficient code in a language other than Python only needs to mention it once, and the system understands. Shopping for groceries for a large household is simplified, with the model considering the need for 6 servings in the shopping list.
Plugins
Users can enhance their experience with plugins by adding pertinent instructions. For instance, if a user indicates their city within the instructions and employs a plugin for restaurant reservations, the model may incorporate that city when interfacing with the plugin.
Beta
During its beta phase, ChatGPT might not flawlessly interpret all custom instructions. There will be instances where it may disregard certain instructions or implement them undesirably.
Safety
Safety protocols have been adjusted to consider the innovative ways users can guide the model. For instance, the Moderation API ensures that instructions contravening OpenAI’s Usage Policies aren’t saved. Moreover, the model has the discretion to decline or bypass directives that could lead to content violating these policies.
Privacy
OpenAI might utilize custom instructions to refine model efficacy for its userbase, but this feature can be toggled off in the data control settings. Like all ChatGPT interactions, measures are in place to expunge personal identifiers from custom instructions prior to their use for performance enhancement. Further insights into how conversations are harnessed for model improvement and related user options are available in the Help Center.
Test it Out
Plus plan subscribers have the opportunity to engage with custom instructions immediately by joining the beta:
- For web users: Navigate to your profile by clicking on your name → Choose Settings → Go to Beta features → Enroll in Custom instructions. This feature will be accessible from the dropdown after clicking on your name.
- On iOS: Access Settings → Navigate to New Features → Activate Custom instructions. The option will then be available within settings.
Please note that this functionality has not been introduced in the UK and EU as of today.