ChatGPT gets more personal: Now users can give it 'custom instructions'
No one likes repeating themselves. OpenAI realized that.
So they launched ‘Custom Instructions’ for ChatGPT users on Thursday, so they don’t have to write the same instruction prompts to the chatbot every time they interact with it.
Currently in beta phase and available to only Plus users, ChatGPT will give a user the option to prefix their prompt with instructions that the chatbot will remember in its memory all the time.
OpenAI said in a blog post: “We’ve heard your feedback about the friction of starting each ChatGPT conversation afresh.” The new feature will save up a lot of repetitive prompting.
So how does it work?
As per OpenAI’s blog post, there are two questions that the user will have to answer as part of using ‘Custom Instructions.’ The first is ‘What would you like ChatGPT to know about you to provide better responses?’ and the second is ‘How would you like ChatGPT to respond?’
Say for example, if a user is a chef and uses ChatGPT frequently for recipes, then they can answer the first prompt along the lines of: “I am a chef at a Manhattan restaurant.” And for the second question, the user can type in: “When I ask for recipes, give me a variation of 3-4 recipes from the best cooks in the world. And also, make sure the proportions are sized to serve only one person.”
The character limit for typing in a user's response is currently gated at 1500. Custom instructions has generated a fairly positive response on social media.
First time trying 'Custom Instructions' - New ChatGPT-4 update.
— Edward (@forward_notion) July 22, 2023
I added - " ChatGPT acts like a philosophy expert and poet. All responses are written in poetry" as a custom instruction, just to test it out pic.twitter.com/fD25HtQfpr
The latest feature, custom instructions, is absolutely astounding!
— ibrahim uzun (@ibrahimuzn) July 21, 2023
You can simply use "Act as a..." and effortlessly receive the desired response from Chat GPT without revisiting the custom instructions settings.
I've attached the custom instruction I crafted for you to test: pic.twitter.com/mdN9Vi6VbE
OpenAI cognizant of drawbacks
The company said that especially during the try out or beta period, the chatbot won’t always interpret a user’s instructions perfectly. At times, ChatGPT might overlook instructions, or apply them when not intended.
Addressing safety concerns, OpenAI said that their Moderation API is designed to help ensure instructions won't be saved if they violate their Usage Policies. The model can also refuse or ignore instructions that lead to responses that violate our usage policies.
The Verge spoke to Joanne Jang, who is a strategic product manager at OpenAI, who said that when she asked ChatGPT to give her tips on murdering people, the chatbot refused to answer.
Another drawback would be that in case the chef in the example above wants to use ChatGPT for something other than recipes, they will have to either switch off the custom instructions tool or put in a new set of instructions. This could be time consuming.
Available in 22 countries, but not the UK and EU, OpenAI will eventually expand the feature to all users in the coming weeks.