Bing Edwards / Stable Spread
On Thursday, OpenAI announce A new ChatGPT experimental feature that allows users to provide customized instructions that the chatbot will take into account with each submission. The goal is to prevent users from repeating common instructions between chat sessions.
The feature is currently in beta for ChatGPT Plus subscription members, but OpenAI says it will expand availability to all users over the coming weeks. As of this writing, the feature is not yet available in the UK and EU.
The custom help feature works by allowing users to set their individual preferences or requirements that the AI model will take into account when generating responses. Instead of starting each conversation over, ChatGPT can now be instructed to remember specific user preferences across multiple interactions.
In the foreground blog postOpenAI gave examples of how someone might use the new feature: “For example, a teacher drafting a lesson plan no longer has to repeat that they’re teaching 3rd grade science. A developer would prefer efficient code in a language that isn’t Python—they can say it once, and that’s understandable. Buying groceries for a large family becomes easier, with the model accounting for 6 servings on the grocery list.”
Ars Technica
If you want to give it a try, you’ll need to do some digging in the settings. ChatGPT web users should click their name in the bottom left corner, go to Settings, then Experimental Features, and click the switch next to Custom Help. A new Custom Help option will appear in the menu from now on. For those on iOS, tap Settings, then New Features, and switch Custom Help to the on position.
If you don’t see the option in your ChatGPT settings, it means your account area hasn’t gained access to the feature yet.
After activating the Custom Help feature and clicking on its option in Settings, you’ll see two text entry fields: one that asks what ChatGPT should know about you, and one that asks, “How would you like ChatGPT to respond?”
Ars Technica
OpenAI says the first input field is designed to gather contextual information about the user that can help the form give better responses. It is suggested to write information about the user’s profession, preferences or any other information that may be useful to the form. For example, “I work on science education programs for third graders.”
The second field is used to collect details about how users would like the AI model to respond. This allows users to direct or format the form output. For example, users can instruct ChatGPT to present the information in a tabular format, specifying the pros and cons of the options in the result.
While testing the feature, Ars found that the Personalized Help feature improved results in a similar way to presenting the same information at a prompt. In this way, it appears to work as a quick extension, although we’re not sure if the characters (1500 in each prompt) count against the general context window. The context window is currently limited to 4096 tokens (parts of a word), and it limits the total short-term memory, so to speak, of the current conversation. We’ve reached out to OpenAI for clarification.
Custom instructions also work with plugins, OpenAI says. For example, “If you specify the city where you live in your instructions and use a plugin that helps you make restaurant reservations, the form may include your city when it calls the plugin”. But the company also cautions in the ChatGPT interface that custom instructions, such as prompts, are shared with plug-in providers.
The company says that while personalized instructions will be applied to all future chats once enabled, the new feature may not work perfectly during beta testing. ChatGPT may omit instructions or apply them in contexts for which they were not intended. For now, OpenAI hopes to collect feedback about the feature from users during the beta period.
“Typical beer trailblazer. Hipster-friendly web buff. Certified alcohol fanatic. Internetaholic. Infuriatingly humble zombie lover.”