ChatGPTis gaining new features at a breakneck clip, but the addition of “custom instructions” to ChatGPT’s beta features might turn out to be the biggest game-changer yet for the chatbot—especially if you want a personalized AI assistant. Here’s what it does and how to use it.
What Are Custom Instructions?
As a Large Language Model, ChatGPT can understand instructions in natural language and act in accordance with those instructions. So, at the start of the conversation, you may specify the rules of engagement, and give ChatGPT the information it needs to do the work you’re asking of it.
With custom instructions, you may embed information and instructions into ChatGPT that persists across all chats. The chatbot will always take the contents of its custom instructions into account in every conversation, unless you specifically tell it not to.
Think of it like Robocop’sPrime Directivesor Asimov’sThree Laws of Robotics, except you get to make the rules. The number of reasons this is useful is limitless, but for starters you can make sure that ChatGPT always knows who you are, what you do for a living, andwhat you need it to do. You can specify what personality it should have, or put up safety barriers that prevent it from doing specific things. Now that you have a good idea of what custom instructions are, we’ll look at how to activate them.
How to Activate and Use ChatGPT Custom Instructions
We’ll assume that you’ve already created a ChatGPT account, and that you’ve signed up for a ChatGPT Plus subscription. Once you’re logged in and at the chat interface, here’s what to do:
Now when you start a new chat, ChatGPT will take your custom instructions into account. If you want to double-check, you can prompt it to tell you what those instructions are before you proceed.
How Well Do Custom Instructions Work?
In my testing, custom instructions have worked remarkably well, and now ChatGPT has some continuity between chats, since it knows who I am and what I want without being told every time. That being said, I have caught it slipping up every now and then, such as mentioning its knowledge cut-off date despite being instructed not to. The feature is, after all, still in beta. So as with anything ChatGPT-related, you’ll always have todouble check its outputsto ensure they match what you asked for.
