Sam Altman’s company opens a marketplace for specialized AI programs. With this, Open AI wants to secure its dominance.
Chat-GPT has many areas of application. Customization is easier with GPT.
Annick Ramp
Back in November, Open AI announced that it wanted to open a kind of app store for GPT, small, tailor-made AI programs. It should happen in the next few days.
What is a GPT?
GPT is the name that Open AI came up with for mini-programs based on its linguistic artificial intelligence (AI). As with Chat-GPT, it is operated via typed commands and, in some cases, by uploading files.
In contrast to a general request to Chat-GPT, the GPT provides the AI with certain framework conditions that it must take into account when processing it. For example, that a question should only be answered using certain documents or in a specific form.
There are specialized GPTs for summarizing scientific papers, generating recipes, explaining board games or automatically creating a presentation on a given topic.
All of this would also be possible directly via Chat-GPT. But it is often time-consuming to achieve the desired result. A new job title has even been created for people who get the best out of AI with tailored requests (prompts), namely prompt engineer.
GPTs already make the prompt engineer partially unnecessary. These programs also enable laypeople to use generative AI for specialized tasks.
What can you use GPT for?
There are countless GPTs on all sorts of topics.
One, for example, is used to generate images that look like they come from a documentary. The following two images show the difference: The first image was created when GPT-4 was asked for an image of the 1920 Rome earthquake. The second arose when the same request was made to the specialized GPT. This automatically improves the query to the image-generating AI.
Result of the query to GPT4 – a surreal image with two coliseums.
OpenAI
The result of the specialized GPT looks a lot more realistic.
OpenAI
Consensus, a search engine that uses AI to search scientific papers to answer questions and support them with sources, also created a GPT. This means that questions that you ask the chatbot are forwarded to the research database in the background, so that Chat-GPT cites research papers in the answer to a question. In our test, however, the quality of the answers was still not perfect.
Who can create GPT?
You don’t have to be a programmer to create a GPT. Open AI provides paying users with an editor called “GPT Builder” that provides step-by-step instructions. This means that even laypeople can create a simple GPT program.
In order to build more complex GPTs, you should already be well versed in generative AI and writing prompts. What is particularly interesting is the possibility of allowing GPT to interact with external programs and documents: you can, for example, integrate it into your email program so that websites can query or access a database.
This works via open interfaces (APIs). Using these interfaces correctly requires some expertise. And because the AI interacts with external content and programs through it, errors can have serious consequences. An automatic email responder could be tricked by attackers into revealing the contents of emails. If the GPT can modify a database, errors can mean important data is deleted. The more powerful the program, the more careful you should be.
What is the GPT Store?
You can build a GPT for yourself, but you can also share it with other users, such as work colleagues. Or you can make it public to anyone who is interested and has a paid account with Open AI. At the moment you simply share the relevant link. However, this makes it difficult to find good GPT. A transparent overview with reviews from other users is still missing.
This is exactly what a GPT store can do. It is intended to make sharing GPT easier and clearer. The details are still unclear.
In the Apple App Store, programs are checked before they come online. Sometimes users have to pay – although Apple, as the platform operator, receives a generous commission. Google Play works the same way. It is not yet known how exactly the business relationships at the GPT store will be regulated.
What does Open AI want to achieve?
If there are many interesting programs to be found in the GPT store, this makes a subscription to Open AI more attractive. That alone would be reason enough to open the store. Additionally, it is more attractive for developers to invent a GPT app if there is a clear and meritocratic marketplace where their program can be found and perhaps even make money.
The GPT store could also harm some people. Open AI is thus competing with some of its own customers: all those who used Open AI in the background to offer a specific service, such as an automatic presentation creator or a tool for answering emails. The customers of these startups do not need a contract with Open AI, everything goes through the intermediary company.
This type of middleman can better control Open AI by offering their services as GPT than by building external products with it. With the marketplace, Open AI is also expanding its power and control in the AI value chain.
The GPT store is also a possible advantage over other large AI manufacturers. The chatbots from Google, Anthropic or the European AI models from Mistral or Aleph Alpha are catching up in terms of quality. Open AI can either continue to differentiate itself through new research and increasing quality. But nobody knows when this path will reach its limits. So Open AI is also trying to shine on other levels: through ease of use and an attractive ecosystem. The GPT store could help with both.