OpenAI recently made a rather interesting announcement - the introduction of custom instructions for their large language model, ChatGPT.
It's an innovation that looks to offer a more tailored and personable experience for users, one that really got me pondering about the exciting potentials in the realm of conversational AI.
Subscribe or follow me on Twitter for more content like this!
The timing of the announcement is interesting as it comes on the heels of a recent finding that shows GPT-4 performance has drifted downward in some areas over time (though I suspect this is a coincidence).
Let's look at what the OpenAI team announced...
Now, the concept is simple but highly significant. You, as the user, can set preferences or requirements that you want ChatGPT to consider while generating responses. This means that you no longer have to repeat your preferences or relevant information in every conversation.
ChatGPT will keep them in mind for all future conversations.
Consider this: you're a teacher crafting a lesson plan for third-grade science or a developer preferring efficient code in a language other than Python. Once you've communicated this to ChatGPT, it keeps it in mind.
Even grocery shopping for a big family becomes more convenient, as the model accounts for the size of your family while drafting the grocery list based on what you tell it. Or, if you want to only get complete code and not have to hear a ton of explanation, well, you can set that too.
This is all quite fascinating as I can see several use cases already for my own workflows!
The key takeaway here is that you can tell ChatGPT about your needs, preferences, or context, and it will endeavor to use that information to generate better, more fitting responses each time.
As an example, let's say you work on science education programs for third-grade students and want responses presented in a table format, outlining the pros and cons of each option.
ChatGPT will attempt to tailor its output to your specifications, making it easier for you to compare and make decisions. These preferences will be maintained over time too, so you don't have to repeat yourself.
Custom instructions also come into play when you use plugins, allowing you to share pertinent information. For instance, if you specify your city of residence in your instructions and use a plugin for restaurant reservations, the model might include your city when it activates the plugin.
However, we must remember that the custom instructions feature is still in beta, and ChatGPT may not always interpret these instructions perfectly. It might occasionally overlook instructions or apply them unintentionally.
OpenAI has also made sure to take care of safety and privacy aspects. They've tweaked their safety measures to account for the new ways users can instruct the model. The model can refuse or ignore instructions that may lead to responses that violate OpenAI's usage policies.
The instructions you provide may be used to improve the model's performance, but you have the control to disable this. OpenAI takes steps to anonymize data and remove personal identifiers before using the information to enhance the model's performance.
Plus, users can follow the steps below to enable custom instructions:
This is not available to UK and EU users currently.
So, Plus users can start using custom instructions by opting into the beta, and this feature will be available to all users in the coming weeks. It seems like we're taking another leap forward in making AI an even more personalized tool, but it's still early days, and I'm keen to see how this development unfolds.
It's time we embraced this evolving technology and saw firsthand the changes it brings to the world of conversational AI. Exciting times are ahead!
Subscribe or follow me on Twitter for more content like this!
Also published here