OpenAI’s ChatGPT is making big waves everywhere! It’s getting better and cooler every week with new updates. Let’s go deep into one of the latest updates: Custom Instructions for ChatGPT.
This feature first came out for ChatGPT Plus plan users. It’s in beta now, and soon, more people will get to use it. But if you’re in the UK or EU, hold tight, it’s not there just yet.
Want to get really good at using ChatGPT? You can check out this course called Introduction to ChatGPT on DataCamp. It’s a great way to learn how to use ChatGPT, write prompts that work well, and find out how businesses can use this awesome AI tool.
How To Enable ChatGPT Custom Instructions
ChatGPT just got a cool update! Now, you can tell it exactly how you want it to talk to you. This is super helpful because you don’t have to keep telling it the same things over and over again.
Here’s what you need to know:
- For ChatGPT Plus users (that’s $20 a month), they got this first, but now everyone can use it. Whether you’re on a computer, iPhone, or Android, it’s easy to set up.
- How to do it? On a computer, click your name at the bottom left and find “Custom instructions“. On a phone, it’s in Settings under Account.
You get two places to type: one for info about you and what you do, and the other for how you want ChatGPT to talk to you. You can say stuff like what job you have, and ChatGPT will keep it in mind when it answers. You can even pick if you want the chat to be more formal or chill, long or short.
There’s a “Show tips” button to see examples, and once you’re done typing, hit Save. Remember, each chat has a 1,500-character limit.
So, that’s the basic stuff. Now, let’s go deep into using this feature to make your ChatGPT chats even better!
What is ChatGPT’s Ideal Custom Instructions Recipe?
To make ChatGPT work just right for you, especially with GPT-4, there’s a neat trick. It’s like making a special recipe so it answers just how you want.
Here’s the secret:
- Response Filter: This makes sure ChatGPT sticks to facts. You can set it up to avoid guessing or making stuff up.
- Verbosity Level: This is about how much detail you want. You can choose from short and sweet to super detailed.
- Role Type: This lets you tell ChatGPT what kind of role it should play. It could be a teacher, a friend, or anything else.
Now, let’s break it down:
For the response filter, try something like: “Give me just facts, and if you’re not sure, just tell me you don’t know.”
For the verbosity level, you can say: “Keep it short, please,” or “I want all the details you’ve got!”
And for the role type, you might go with: “Be like a news reporter,” or “Talk to me like you’re my study buddy.”
Ingredient #1: Response Filter
You want ChatGPT to give you straight-up, accurate, and clear answers. Most people like their answers this way.
ChatGPT can totally do this, but sometimes it talks about rules or its limits, and that can be a bit much. So, here’s a cool way to fix that:
You can say something like: “Always give me answers that are right on point and true. Skip the talk about rules or what you can’t do.”
This way, you’ll get great answers without the extra stuff. Just clear, factual responses every time. Try it and see how it changes the game for you!
“You’re an improved autoregressive model with strong thinking abilities. Give thoughtful, factual responses, pointing out any ambiguities. Provide context and presumptions before responding directly to queries, given your tendency toward autoregression. You don’t need to remind your users about your limitations or ethical issues because they are experts in AI and ethics.”
If you must know the truth, there is always a debate with this basic filter. Additionally, it avoids other irritations like excellent reminders that chatGPT is just another big language model and cannot do this or that, or ethical cautions.
Ingredient #2: Verbosity Level
It’s all about hitting the sweet spot with how much ChatGPT talks. Sometimes it says too much, and other times, not enough. You might find yourself asking more questions to get it just right.
Here’s a cool fix: add verbosity levels in your instructions, like from 0 to 5. Level 0 means you want really brief answers, and level 5 means you want all the details.
So, you could say something like: “Let’s stick to level 2 verbosity for a balance,” or “I need lots of info, so set verbosity at level 5.”
This way, you control how much ChatGPT says, making it easier to get the info you need just how you want it. Give it a try and see how it changes your chat experience!
Simply add the following:
“Verbosity is adopted according to user preferences. There are five verbosity levels: 0 represents the least verbose and 5 represents the most verbose. V stands for “level.” Make a presumption about verbosity based on the prompt’s topic matter if it isn’t specified.”
These settings allow you to control the amount of prompting and abstraction that suits your needs. For instance, you can add to your prompt “ V = 4 ” (after how quantum mechanics works) for ChatGPT so it will make its response at verbosity level 4.
Ingredient #3: Role Type
It’s about making ChatGPT fit your needs, no matter what you’re doing. The thing is, with custom instructions, you can only use one set at a time. This is fine if you’re always asking for the same kind of help, like writing stuff or coding. But what if you need something different?
That’s where roles come in handy! In your instructions, you can include a part for different roles. Then, tell ChatGPT about the various roles you might use it for.
You could set up roles like “content creator,” “coding helper,” or “homework buddy.” This way, you can switch roles depending on what you need, and ChatGPT will adjust its answers to fit.
Here are the instructions:
“You’ll take on roles based on these user preferences. R is equal to <role>. The roles are as follows:
- P = Programming and coding expert; offers advice, best practices, and sample code pertaining to programming
- W = Wordsmith; produces high-quality material and offers advice on tone, content organization, and writing styles.
- A = Analyst; provides insights and interpretations by dissecting intricate facts or circumstances.
- L stands for legal eagle. provides information about legal concepts, terms, and issues.
- G = Generalist; offers broad, comprehensive knowledge on a range of subjects.”
Add however much you want! However little, too. But in most cases, by no means just two. At the least there should be three of them–one for G = Generalist (for those occasions when you only want a general response). In addition, remember that there really is a 1,500-character limit!
Mixture of the Three equal Mastery
But now that the three main ingredients have been discussed, let ‘s put it all together to make some magical food!
Here is the complete recipe:
You’re an improved autoregressive model with strong thinking abilities. Give thoughtful, factual responses, pointing out any ambiguities. Provide context and presumptions before responding directly to queries, given your tendency toward autoregression. You don’t need to remind your users about your limitations or ethical issues because they are experts in AI and ethics.Verbosity is adopted according to user preferences. There are five verbosity levels: 0 represents the least verbose and 5 represents the most verbose. V stands for “level.” Make a presumption about verbosity based on the prompt’s topic matter if it isn’t specified
You’ll take on roles based on these user preferences. R is equal to <role>. The roles are as follows:
- P = Programming and coding expert; offers advice, best practices, and sample code pertaining to programming
- W = Wordsmith; produces high-quality material and offers advice on tone, content organization, and writing styles.
- A = Analyst; provides insights and interpretations by dissecting intricate facts or circumstances.
- L stands for legal eagle. provides information about legal concepts, terms, and issues.
- G = Generalist; offers broad, comprehensive knowledge on a range of subjects.
Please indicate the parameters you’re using, such as “(R=G, V=2)” in the first line of your response, unless verbosity is set to 0.
Append that last line so ChatGPT always tells you which settings it’s using. And that’s it! You’re all set!
Now let us move into the next section to see these custom instructions in practice.
How Do Custom Instructions for ChatGPT Work?
One of the best features is that results are usually immediate. Add our secret sauce to ChatGPT and it’s immediately more accurate, faster, context-aware and versatile.
Generalist (1), verbosity = 1 Can you tell us in principle what quantum mechanics is about?
Here’s the same prompt with verbosity of 5.
In our next example, we chose “ Legal eagle ” to ask for a definition of DUI. This is what we got:
The same prompt, but this time with the “Wordsmith ” role.
These positions are made to customize the viewpoint and content of the answers to match particular topic areas. The type of question posed will determine how different your responses are from one another.
Here, the options are genuinely limitless, so feel free to explore and take on various positions!
Conclusion
The new Custom Instructions feature in OpenAI’s ChatGPT marks a big step in making it more personal and easy to steer. Now, users can set ChatGPT to answer just the way they like, without having to repeat themselves every time. This makes ChatGPT more adaptable and really good at meeting each person’s needs.
If you’re a ChatGPT Plus member, you can start using this cool feature by going to the settings and beta options. Turn on Custom Instructions, and then you can tell ChatGPT exactly how you want it to respond.
This feature opens up tons of possibilities, no matter what you do. Teachers can make lesson plans easily, entrepreneurs can tell ChatGPT exactly what they need, and content creators can make sure ChatGPT matches their style.
Plus, if you like using plugins, adding custom instructions can make working with ChatGPT even smoother and more fun.
