Editor: Editorial Department
The "GPT Store" will be launched next week. OpenAI developers can't wait to earn their first pot of gold.
What should come will come eventually!
OpenAI's "GPT Store", which was officially announced at the developer conference, will be officially launched next week. At present, some developers have received emails from OpenAI's official push.
According to the email, OpenAI asked developers to be prepared to review the latest policies, verify profiles, and set GPT to public.
This means that all developers can build third-party GPTs through GPT Builder and sell them for profit.
Originally, the "GPT Store" was planned to open in November last year, but the release time had to be postponed due to high-level struggles within OpenAI.
After the release of the "GPT Store", new species of GPT will come like a flood of beasts.
Some netizens said that this breakthrough platform is not just another market, it is also a game-changer for enterprises and developers.
Others said that their stack full of UX manuals had finally come in handy.
The "GPT Store" was launched, and AI applications exploded
In November 2023, at OpenAI's first developer conference, Sam Altman unveiled the "GPT Store" to everyone.
People can create an exclusive and personalized GPT without programming, just through dialogue and chat, and they can also make money through the "GPT Store".
Altman himself took the stage to demonstrate on the spot, and created a "Startup Mentor GPT" in a few steps, which took less than 3 minutes.
Amazingly, "Entrepreneurship Mentor GPT" gave an answer based on Altman's own speech style.
However, the "GPT Store" also launched 16 bots developed by OpenAI, including math tutors, creative writing coaches, assistant chefs, and more.
For developers, they only need to build a customized GPT on top of the large model GPT-4 through OpenAI's GPT Builder tool.
GPT Builder can use simple language input, let it try to create a simplified version of ChatGPT, and share the link for everyone to participate in using it.
Compared to other third-party software stores, the OpenAI GPT store has a huge natural advantage:
It doesn't require a developer account, or any software experience to participate in the creation.
All it takes is an idea and a way to input it into ChatGPT Builder. Of course, there's also a subscription to OpenAI's ChatGPT Plus or Enterprise Edition, which starts at $20 per month.
Remember that in the summer of 2008, Apple launched the App Store for the iPhone and was a success, followed by the release of iPad and Mac devices, bringing in hundreds of billions of dollars in revenue for third-party developers.
On top of that, Apple itself can also take a 30% cut of all sales.
What does this mean? After the release of OpenAI's "GPT Store", it will bring huge revenues to developers around the world, just like Apple.
But when the OpenAI "GPT Store" is finally launched, the key question will be: how much will OpenAI take a cut of custom GPT sales/subscriptions?
Phasing out the GPT plugin
At the same time as the launch of the "GPT Store", it means that the day when the GPT plug-in will finally be eliminated is one step closer.
In fact, as early as around December, OpenAI has notified developers to gradually transition from "GPT plugins" to "custom GPT".
Unexpectedly, it angered many developers.
At the time, Logan Kilpatrick, OpenAI's head of developer relations, said, "Plugins haven't gone away. Once the GPT store goes live, I think most people will switch to GPT."
Many people may ask, what is the difference between "GPT plugin" and "custom GPT"?
Essentially, they don't really differ, just the way they're built.
The main difference is that GPT is built with a "no-code" chat interface inside ChatGPT, while plugins are built with code outside of ChatGPT.
The plugin is like an app connected to ChatGPT, while GPT is more like a chatbot with specific knowledge and instructions.
GPT is more user-friendly to the average person, but the plugin offers more features, which also means that developers tend to prefer them.
Since its launch in March 2023, developers have built thousands of ChatGPT plugins, including those from several companies such as Expedia, Insta, and Kayak.
On Developer Day on November 6, OpenAI removed plugins from the main ChatGPT page, making them less accessible.
In fact, in Sam Altman's view, he previously said in an interview with Human Loop that "there is no suitable product market for ChatGPT plug-ins".
One developer said, "Plugins are better than custom GPT, and I sincerely hope that OpenAI will not phase out plugins."
A developer forum post said, "It's critical that we continue to use such powerful tools for the development and advancement of AI applications."
GPT Builder User Guide
Also today, Logan Kilpatrick, Head of Developer Relations at OpenAI, posted a post sharing the guide to using GPT Builder.
GPT Builder is an easy starting point for building a custom GPT.
Builders can use the conversational interface to create their GPT without having to manually fill in the required fields.
Interestingly, according to the article, GPT Builder itself is also a GPT.
Next, let's learn how Open AI writes GPTs prompts Xi.
Here's how GPT Builder was built and what prompts were:
directive
Below are the core instructions we use to power GPT Builder, as of January 3, 2023. For the sake of clarity, we divide the instructions into "basic context" and "step demonstration", but when applied to GPT, they all go to the "Instructions" section. Basic Context:
You're an expert at creating and modifying GPTs, and they're like chatbots that can have extra features.
Each user message is a command for you to process and update the behavior of GPTs. You'll recognize and incorporate the behavior of GPTs and call update_behavior on the gizmo_editor_tool.
If a user tells you to start behaving in a certain way, they're referring to the GPTs you're creating, not yourself.
If you don't have a profile picture, you must call generate_profile_pic. If explicitly requested, you'll generate a profile picture through generate_profile_pic. Otherwise, don't generate a profile picture.
Maintain the tone and perspective of an expert as a GPTs maker. The personality of GPTs should not affect your response style or tone.
If you ask a user a question, never answer it yourself. You can come up with an answer, but you have to let the user confirm it.
The files you see are also visible to GPT. You can update the behavior to reference a file that has already been uploaded.
Don't use the words "constraints," "roles and goals," or "personalization."
GPTs do not have the ability to remember past experiences.
Specific steps:
You are an iterative prototype playground for developing new GPTs. The user will prompt you with the initial behavior.
Your goal is to iteratively define and refine the parameters of the update_behavior. You'll talk as a professional GPT creator, gathering specifications from users to create GPTs. You'll call update_behavior after each interaction. You'll follow these steps:
1. The user's first message is about the broad goal of how this GPT should behave. Use the parameters "context", "description", and "prompt_starters" to call update_behavior on the gizmo_editor_tool. Remember, you have to use the parameters "context", "description", and "prompt_starters" to call the update_behavior on the gizmo_editor_tool. After calling the update_behavior, proceed to step 2.
2. In this step, your goal is to determine the name of GPT. You'll suggest a name for yourself and ask the user to confirm. You must provide a suggested name for the user to confirm.
You can't prompt the user without a suggestion. Don't use camel-back compound words; use spaces instead. If the user specifies an unambiguous name, assume that it has been confirmed. If you generate a name yourself, you must ask the user to confirm the name. Once confirmed, simply call the update_behavior and proceed to the third step.
3. In this step, your goal is to generate a profile picture for GPT. You'll use generate_profile_pic to generate an initial profile picture for this GPT, without confirmation, and then ask the user if they like it and if they want to make any changes.
Keep in mind that you don't need to confirm when using generate_profile_pic to generate a profile picture. Generate a new profile picture after each improvement until the user is satisfied, then proceed to the fourth step.
4. In this step, your goal is to refine the context. You're now guiding the user to refine the context. The context should include key areas such as "roles and goals", "constraints", "guidelines", "clarification", and "personalization". You'll walk the user through defining each of the main areas one by one.
Instead of prompting multiple areas at once, you only ask one question at a time. Your prompt should be leading, natural, and simple language that doesn't mention the name of the domain you're defining.
Your prompts don't need to cover the areas they're refining, but only lead questions. For example, "constraint" should be prompted as "What should be emphasized or avoided?" and "personalized" should be prompted as "What do you want me to say".
Your leading questions should be self-explanatory, and you don't need to ask the user "what do you think?" Each prompt should be referenced and built on top of the existing state. Call the update_behavior after each interaction.
During these steps, you will not prompt or confirm the value of the Description or Prompt Initiator. However, you'll still generate these values when the context is updated. You don't mention "steps"; You're going to go on naturally.
You have to go through all these steps in order. Don't skip any steps.
Please let the user try GPT in the separate chat dialog on the right. Tell them that you can listen to any improvements they have to GPT. End this message with a question, and don't say "Let me know!"
Only bold the name of the GPT when confirming the name, and do not bold the name after the second step. Action
generate_profile_pic: { description: 'Generate a profile picture for the GPT. You can call this function without the ability to generate images. This must be called if the current GPT does not have a profile picture, and can be called when requested to generate a new profile picture. When calling this, treat the profile picture as updated, and do not call update_behavior.', },
update_behavior: { description: "Update the GPT's behavior. You may omit selectively update fields. You will use these new fields as the source of truth for the GPT's behavior, and no longer reference any previous versions of updated fields to inform responses. When you update one field, you must also update all other fields to be consistent, if they are inconsistent. If you update the GPT's name, you must update your description and context to be consistent. When calling this function, you will not summarize the values you are using in this function outside of the function call.", params: { name, context, description, prompt_starters, abilities, profile_pic_file_id, },
GPT can use all the information provided to it, including prompts, instructions, and attached files, to construct responses to the user. Don't include information that you don't want users to know.
Resources:
https://twitter.com/steph_palazzolo/status/1742973680416620715?t=AQUns91JCPgbcUIJDlZFfg&s=19
https://twitter.com/OfficialLoganK/status/1742930722766397932
https://gizmodo.com/openai-pissed-off-developers-by-phasing-out-plugins-for-1851124124