In a recent interview with Sam Altman, there was a dramatic scene.
Altman looked up at the sky, touched his finger over his Adam's apple, thought for a moment and said, "I think it's a little bad." 」
Altman is undoubtedly a good CEO who has raised expectations for GPT-5, which is expected to be released in the middle of the year.
But I don't know if he remembered the GPT Store in his mind at this time, which was also officially announced by him.
GPT Store, which was launched in January, has gradually become unconcerned after more than 2 months, and only enjoys the taste of being the protagonist for 15 minutes.
Panning for gold from the GPT Store is like opening a blind box
GPTs in the GPT Store are custom versions of ChatGPT.
We don't need to write code, we can use natural language to converse, and we can make a ChatGPT with unique skills, such as a "digitally proficient media editor", and more professional players can upload files or call third-party APIs.
When the GPT Store first launched, there were already more than 3 million GPTs, but they were only available to ChatGPT Plus paid users.
I don't know if it's because there are too many white prostitutes waiting for GPT-4 to be free, or if the GPT Store itself is not angry, and the use of GPTs is a bit bleak.
The Information report mentions that a developer analyzed more than 36,000 GPTs, and about 5% of them had 150 to 500 active users per day, but the vast majority had only 1 to 2 users per day.
"It feels like OpenAI has given up on GPT Store. The developer complained.
Anyone who finds the reason from himself in advance, in terms of user experience, compared with January, GPT Store has improved, but not much.
The interface of the GPT Store still only has a few categories such as writing, coding, and productivity, and there are more than a dozen recommended GPTs under each category.
But the user's needs are trivial and vague, and more than 2 months later, ChatGPT still can't provide an interface that allows us to describe the needs in natural language and then call up the required GPT.
However, some developers have found this gap and launched "GPT Finder" and other GPT to find GPT, and find relevant GPT links on Google, Bing, X, third-party GPT navigation websites and other platforms through the web search function.
If the mountain doesn't come, I'll go to the mountain. When I swipe the information flow such as instant and Xiaohongshu, I occasionally see a self-recommendation or recommendation from a certain GPT, and if the function hits the pain point, I will also use it.
So, the process of finding the GPT you want is somewhat random. However, finding it is only the beginning, whether GPT is useful or not, the probability is similar to that of opening a blind box.
Most GPTs are very simple and well-targeted: save you time searching and finding, drive traffic to your website and services, register your email address, and attract your valuable subscriptions.
For example, when you ask Canva, a well-known design tool, to help you make a greeting card, it provides a template that invites you to click on the image and jump to the website to edit it.
Perhaps that's why Canva has millions of conversations and tens of thousands of ratings, but it's only 3.2 out of 5.
GPTs that don't need to jump don't necessarily save time and effort.
I wanted to see if GPTs could help me make annoying Powerpoints, so I searched for the keyword "Powerpoint" and found one of the most frequently used ones.
However, it can only be "on paper", giving an outline based on my topic, without data, pictures and tables, only abstract textual guidance.
What's worse is that it is carefully used in the wrong place, reminding me to use large fonts and contrasting colors, and even suggesting that I can add a "question mark" picture to a certain page, indicating that it is time to interact with the audience, not like an assistant, but like Party A who is easy to have a flash of inspiration.
In some scenes, you don't even need to invite GPTs out of the mountain, and there is a feeling of "killing chickens with a knife", which is not GPT, but the energy we look for.
For example, if you find a GPT that specifically summarizes the book and then talk to it, it takes more time, and the effect is not necessarily better than GPT-4.
▲ The picture above shows a GPT, and the picture below shows GPT-4.
Of course, there are also particularly useful and indispensable GPTs, such as translating scientific articles and writing prompts for Wensheng Diagram AI, but these are not what I found in the GPT Store, they are all spread by word of mouth on social media.
In the past two months, the most interesting new feature of GPTs should be "@", similar to @ someone in a WeChat group chat, typing @ in the chat box can summon multiple GPTs in the same chat interface, allowing users to be bosses and AI employees.
However, the scope of @ is also limited to known GPTs, which are either recently used or have been added to the sidebar and require you to actively choose, and there is no good thing for you to input and help you pick GPT.
"It's a pity to eat tastelessly, it's a pity to discard it", finding good GPTs is love, forgetting to use GPTs is duty.
The GPT developers who generate electricity for love are in a light mood
When launching the GPT Store, OpenAI promised that developers could make money using GPT, starting with the United States, and more details should be seen in the first quarter of this year.
However, many GPT developers don't do it to make money, they use GPTs as a minimum viable product (MVP), attract seed users, test the product's viability and market demand, and then iterate on the product.
At the moment, however, GPTs doesn't give much feedback to most developers.
Most obviously, restricting GPTs to ChatGPT paying users directly limits the number of users.
In November 2023, ChatGPT has reached 100 million weekly active users. Lively is lively, and the payment rate is not as high as imagined. Some estimates suggest that ChatGPT has between 5 million and 8 million paying users.
The traffic ceiling is visible to the naked eye, and at the same time, it is difficult for later developers to overtake in corners, because the GPT Store will hardly give traffic and feedback to newly launched GPTs.
GPT Store follows the Matthew effect, GPT with higher usage in the early stage has higher weighting, and when we search, the most conspicuous indicator of GPTs is the number of uses.
Such an elimination mechanism is simple and crude for OpenAI, but those individual developers who can't make the list have less incentive to generate electricity for love. The absence of fresh blood is not conducive to the long-term development of GPT Store.
In addition, because the prompt words are easy to leak, the plagiarism of GPTs is still a problem, and OpenAI has not been able to solve it so far.
The problem is that even if you are really at the top, the flow is limited. GPT, a text escape game created by a developer, appeared at the top of the GPT Store for 2 weeks, adding only 50,000 new conversations, far less than he expected.
With a smaller user base, developers expect higher quality feedback.
From a user's point of view, GPT's "first impressions" are richer than before, and before you start chatting with GPT, you can see information such as profile, rating, genre ranking, number of conversations, and other GPT from the same developer.
When we talk to a GPT for about 6 rounds, the evaluation portal will actively pop up in the chat interface, allowing you to score 1 to 5 stars like a scoring movie on Douban.
Another evaluation method is more hidden, click on GPT's name to find the evaluation entrance, not only can you make a star, but you can even leave a message to the developer and send it to the other party's mailbox.
However, the reviews received by developers can be rough.
The Information report mentions that the GPT Store only provides developers with a fraction of the data on user satisfaction, engagement, and more. At the same time, OpenAI resets user IDs every 24 hours, and developers can't track how many duplicate users there are.
Instead, many developers share and promote their GPT and receive feedback on communities such as Reddit.
As of now, individual developers have not been able to get a piece of the pie or more useful feedback, and their attitude towards GPT Store has become light, in line with the mental state of today's young people.
GPTs in place of plugins, waiting to be replaced
Recently, OpenAI announced one thing in a low-key way: ChatGPT Plugins (plugins) will be completely taken offline from April 9.
The plugin was pronounced dead in December last year, only to be buried in April. At the time, OpenAI announced that it would replace the plugin with GPTs, saying that GPTs had learned from the plugin.
It can be said that the plug-in is the predecessor of GPTs, and it was also slapped to death by GPTs on the beach. The transition from plugins to GPTs reflects OpenAI's understanding of applications in the AI era.
The plugin is built using code outside of ChatGPT and connects to ChatGPT via API.
GPTs are created through ChatGPT's internal no-code chat interface, and its "Actions" are plug-ins, which can call one or more APIs to enable more customized actions and interact with external data or the real world.
In terms of threshold, GPTs are of course more friendly, users who don't know how to code can get started, and excellent developers know better how to write prompts, allowing GPTs to search the Internet, call APIs, and perform semantic searches on uploaded knowledge documents.
Last June, Sam Altman said of the plug-in: Plug-ins are not suitable for the product market.
He explained that many plugin developers seem to be integrating their own applications into ChatGPT, but in fact they want to integrate ChatGPT into their own applications.
To some extent, plug-ins are still developing applications in the AI era according to traditional ideas, which is a bit out of place. During the pre-testing of the plugin, the user experience was also very poor, and the API often crashed when calling it.
In contrast, GPTs are lighter, more stable, and more in line with our imagination of "Internet portals", changing the process of retrieving, acquiring, and processing information without having to switch between different applications.
GPTs have replaced plugins, and perhaps in the future it will be replaced as well.
Steve Jobs famously said, "People don't know what they want until you show it to them." He also gave the example of Ford, where if you ask a customer what they want, they may just want a faster horse.
Compared with ChatGPT, GPTs are also "faster horses", not a revolutionary change, its significance lies in the fact that on top of the parameters of the model, it is combined with the user's daily life to solve more specific problems, but many times it is not so useful.
As OpenAI itself puts it, "The GPT Store expands the positive impact and creative uses of ChatGPT. 」
At the same time, the GPT Store may be a good testing ground for developers to discover the needs of users, and for ordinary people to find a sense of creativity.
Emilia David, a reporter at The Verge, made a synonym for GPT herself, which she used to search through Google or look up on sites like Thesaurus, saved her 30 seconds of time and didn't have to watch ads.
She also found that GPTs weren't fully integrated into her workflow because it was based on ChatGPT Plus, not the document she wrote about, and if she hadn't paid for ChatGPT, there would be nothing to Google search back.
GPTs' slightly awkward position illustrates the evolution of our needs for AI applications. It is like a transitional container that carries the experience that can be carried forward.
A week ago, the world's first fully autonomous AI programmer was introduced, and the AI Agent craze was once again in Silicon Valley.
The Information broke the news in February this year that OpenAI is also developing its own AI Agent, which will take over users' devices and automate complex tasks, which is not the same thing as ChatGPT.
Chatting with AI for a few days and getting the job done is the promotional tone of AI Agents. Before this day comes, GPTs, which make users frown and smile bitterly, may be imperfect and contain more possibilities for applications in the AI era.