Update: Some users have noticed that the custom instructions option is currently unavailable. We are investigating this matter, and while we work on resolving it, please be aware that the function may not be accessible at the moment.
OpenAI has recently introduced a new feature for ChatGPT users, allowing them to customize instruction prompts. This means that users no longer have to repeatedly provide the same instructions when interacting with the chatbot. Instead, they can now include specific inputs such as “Limit the response to 1,000 words” or “Keep the tone formal” to guide ChatGPT’s responses.
According to the company, this feature empowers users to share any relevant information they want ChatGPT to consider in its responses. As an example, a teacher can indicate their specialization in fourth-grade math, while a developer can specify their preferred programming language when requesting suggestions. Even personal details like family size can be specified, enabling ChatGPT to offer tailored responses regarding meals, grocery shopping, and vacation planning, among other topics.
Users can now utilize custom instructions with ChatGPT, a valuable addition for those who frequently need to establish specific contexts during their interactions. Although users can already specify details while chatting, custom instructions offer a more convenient way to set consistent contexts.
This feature is also compatible with plug-ins, streamlining the process of suggesting restaurants or flights based on the user’s location.
OpenAI has clarified that the custom instructions feature is accessible to Plus plan users, but regrettably, it won’t be available to users residing in the EU and the U.K. It is currently in the beta phase.
To try out this exciting feature on the web, users can click on their name and navigate to Settings > Beta features > Opt into Custom instructions. On iOS devices, access can be found through Settings > New Features > Turn on Custom Instructions.
Significantly, OpenAI has clarified that the data provided for customizing responses will be utilized to train its API models, allowing them to adeptly adapt to a wide array of instructions.
OpenAI stated that data from the utilization of custom instructions will contribute to enhancing model performance, ensuring that the model learns to tailor responses to user instructions effectively, striking the right balance without going overboard. Nevertheless, users retain the option to opt out of this setting by accessing their data control settings.
The company has been conducting trials of this feature with selected users, as highlighted by consultant Gavriel Cohen on Twitter. ChatGPT presents users with two distinct boxes to express their chat preferences. In one box, users can provide information about themselves, while in the other, they can specify the desired way to fine-tune the chatbot’s responses to suit their needs.
After users input their responses, the changes will become effective from the next session onward. OpenAI has set a character limit of 1,500 for these customized instructions.
To ensure a secure environment, OpenAI employs its moderation API to scan personalized instructions, carefully examining them for any potential unsafe content. If the instructions are found to lead to harmful or hateful responses from ChatGPT, the system may decline to save or acknowledge them, aligning with the company’s policy and ensuring a positive user experience.
Initially, in May, OpenAI released the ChatGPT app exclusively for U.S.-based users on iOS. However, the company expanded its availability to more than 40 countries shortly after the initial launch announcement. In the previous month, OpenAI launched an iPad app with added support for Siri and Shortcuts. Additionally, the company integrated an option that allows users to search the web for answers through Bing directly from the ChatGPT app.
