OpenAI has rolled out a significant feature called custom instructions, allowing users to have more control over the responses from ChatGPT. The introduction of this feature aims to address user feedback about the difficulties of setting new parameters for each ChatGPT conversation. Custom instructions empower users to set their preferences or requirements, which ChatGPT will then consider in generating responses.
The Role of Custom Instructions
A. Addressing User Needs
Custom instructions are designed to align the AI's responses better with the specific needs of individual users. This enhancement is driven by feedback and conversations with users from 22 countries, reflecting a broad understanding of the diverse contexts and unique needs of individuals.
Example: Teachers no longer need to repeat their specific requirements, such as crafting a lesson plan for third-grade science. Similarly, a developer preferring efficient code in a non-Python language only needs to state their preference once.
B. Maintaining Contextual Preferences
With the new update, ChatGPT will retain the user's custom instructions for all future conversations. This means users won't have to reiterate their preferences or information in every conversation. It simplifies the process and makes the tool more user-friendly.
Example: The new feature makes it simpler for users to carry out tasks, like grocery shopping for a large family. Users can provide the model with instructions to account for six servings in a grocery list, and it will remember these preferences for future use.
Integrating Custom Instructions with Plugins
In addition to improving the conversation experience, custom instructions can also enhance the interaction with plugins by sharing pertinent information. If a user specifies certain details, like their city, in their instructions, and uses a plugin to help make restaurant reservations, the model might use the specified location when calling the plugin.
Limitations and Safety Measures
However, it's important to note that during the beta period, ChatGPT may not always interpret custom instructions perfectly. It may occasionally overlook instructions or apply them when not intended. Despite this, OpenAI has implemented safety measures to account for the new ways that users can instruct the model. The Moderation API ensures that instructions won't be saved if they violate the Usage Policies, and the model can refuse or ignore instructions leading to responses that breach the policies.
Privacy and Custom Instructions
OpenAI also addresses privacy concerns by allowing users to disable the usage of their custom instructions to improve model performance. The company takes steps to remove personal identifiers found in custom instructions before using them to enhance model performance.
Accessing the New Feature
Currently, Plus plan users can access custom instructions by opting into the beta for the feature. For web users, they can access this by clicking on their name → Settings → Beta features → opt into Custom instructions. On iOS, users need to go to Settings → New Features → turn on Custom instructions. However, this feature is not yet available in the UK and EU.
The introduction of custom instructions for ChatGPT is a significant stride towards enhancing user experience and improving AI steerability. By incorporating these updates, OpenAI continues to demonstrate its commitment to listening to user feedback and making necessary improvements for a more effective and personalized AI experience. As the tool continues to evolve, it will be intriguing to see the further enhancements OpenAI has in store for ChatGPT.