ChatGPT Announces Custom Instructions

We’re rolling out custom instructions to empower you to tailor ChatGPT to better suit your needs. This feature will be accessible in beta starting today with the Plus plan, with availability expanding to all users over the next few weeks. Custom instructions enable you to incorporate preferences or requirements that you want ChatGPT to consider when generating responses.

Thank you for reading this post, don't forget to subscribe!

Based on feedback about the inconvenience of starting each ChatGPT conversation from scratch, we’ve engaged with users across 22 countries to grasp the crucial role of steerability in enabling our models to effectively adapt to diverse contexts and individual needs.

Moving forward, ChatGPT will take your custom instructions into account in every conversation. The model will integrate these instructions into its responses, eliminating the need for you to repeat preferences or information in each interaction.

For instance, a teacher designing a lesson plan no longer needs to reiterate that they’re teaching 3rd-grade science. Similarly, a developer preferring efficient code in a language other than Python can express it once, and ChatGPT will understand. Even tasks like grocery shopping for a large family become more seamless, with the model considering specifics like needing 6 servings in the grocery list.

Plugins

Incorporating instructions can also enhance your experience with plugins by providing relevant information to the plugins you utilize. For example, if you specify your city in your instructions and use a restaurant reservation plugin, ChatGPT might include your city when calling the plugin.

Beta

During the beta phase, ChatGPT may not always interpret custom instructions flawlessly—it might occasionally overlook instructions or apply them incorrectly.

Safety

We’ve adjusted our safety measures to accommodate the new ways users can instruct the model. For instance, our Moderation API is designed to prevent instructions from being saved if they violate our Usage Policies. Additionally, the model can decline or disregard instructions that lead to responses violating our policies.

Privacy

While we may leverage your custom instructions to enhance model performance, you can opt out of this via your data controls. Similar to ChatGPT conversations, we take steps to remove personal identifiers from custom instructions before using them to enhance model performance. Learn more about how we utilize conversations to improve model performance and your options in our Help Center.