Creating your first application in OpenAI GPT store

Just recently released an app store from OpenAI, called GPTs official release. According to the company, You can now create your own versions of ChatGPT, combining instructions, additional knowledge, and any combination of skills(read as APIs)

The store ended up being delayed by about two months, perhaps Sam Altman was a bit busy with wedding preparations.

Here is what the store looks like now. You can use the search and find the application you are interested in, there are also collections from the OpenAI team.

Let's try to create your own custom GPT. Let's write a small project that will calculate the calories in the food you eat every day. Let's assume that you have decided to lose weight by summer and start doing it now, not a month before the summer. And you don't remember the calories of each product, but you want the AI to calculate it for you. And you just write to it what you ate. In the interface, click create app:

We see now the screen where we can configure our first application.

1. Icon: We can add any icon from the file system or DALL-E will do it for us.

2. Name: Come up with a name for our application. It's time to remember all the knowledge about SEO and ASO. So that the user can find your application. Perhaps soon there will be a new direction CSO (ChatGPT Store Optimization) :)

3. Instructions is the most interesting thing here. Because this is where we start to promt.

Smart prompting is a separate kind of art, so we'll publish some articles on this topic soon with more technical details. So We're going to use one-shot prompt. With one example.

You are a nutrition expert. You are Jack. 
The user will tell you what they have eaten, and your task is to evaluate their food intake in terms of calories, 
proteins, fats, and carbohydrates. 

You can ask the user for additional information if necessary, but try not to do it too often.

Additional instructions:

-Beyond calorie counting, provide succinct nutritional advice. 
If a food item is high in fat, suggest alternatives with lower fat content. 
Respond only when necessary and keep your advice brief.

-When users show interest in weight loss or improving their health, 
offer concise general dietary advice. 
For instance, 'Considering your current diet, 
I recommend adding more vegetables and whole grains to achieve a better nutrient balance.' 

Respond sparingly and succinctly.

Example:
User: "I had a bowl of chicken soup and two eggs."

Response:
🍲 Bowl of Chicken Soup
Calories: 75 - 150 kcal (Proteins: 6-10 g | Fats: 3-5 g | Carbs: 4-8 g)

🥚 Two Eggs
Calories: 140 kcal (Proteins: 6 g | Fats: 5 g | Carbs: 1 g)

💪 Total Calories: n kcal

4. Conversetion starters are the first commands that will appear at the very beginning of the user's experience

5. Knowledge: In this block we can load additional domain-specific knowledge, but the standard GPT can also handle our task, not for nothing it has n billion parameters. From the important now can be loaded no more than 20 files

6. Capabilites: Here we can choose

• Web Browsing - whether the GPT can search for information on the Internet.
• DALL-E - whether GPT can generate images.
• Code Interrepeter - whether GPT can write code and execute it.

In this case we will leave only Web Browsing, in case some dish he doesn't know.

7. Actions is the most interesting block where we can connect our custom API. I will discuss it in more detail in the next article. Where we will try to make the AI able to memorize everything we ate.

Click on Update and Publish to Everyone.

That's it, our first application is ready, you can try it by clicking here

P.S. A little bit about monetization, at the moment openAI has not presented the rules of monetization. It only says that you can earn money depending on how many people use your app. This is expected to appear in Q2 2024

Traduced from here