On Thursday, OpenAI announced plug-ins for ChatGPT, giving a small group of companies a chance to connect their software directly with the service. In theory, this means that you could use ChatGPT to book a flight through Kayak, make and then order from a shopping list with Instacart, or tell your chatbot to search the web for you.
The company also demonstrated ChatGPT running video software to do some light editing:
OpenAI described this rollout, in muted terms, as just another step in its gradual “iterative deployment philosophy.” This is a bit of an understatement, and nobody in tech mistook it as accurate. ChatGPT already has more than 100 million users, according to the company. Now it has an app store. It can act, in some ways, like an operating system. And it has at least one clear business plan that could matter to regular users: to reorient the entire internet, as well as users’ own devices, around chat.
So far, ChatGPT has mostly served as a tech demo — an early way to interact with a new breed of advanced language models that can mimic human communication and attempt a surprising range of tasks: writing, summarizing, explaining, translating. What it couldn’t do, for the most part, was interact with the outside world or even with the rest of the internet. It was relatively siloed — its outputs and ambitions limited by its slightly and intentionally stale training data.
ChatGPT put on a great show and effectively inculcated a sense of potential about LLMs — and itself and its parent company — in its users. It could do a lot of things for you, but it couldn’t do many things on your behalf. It was, literally, just another web app in a tab in your browser. With plug-ins, it now has control over a browser, through which it can act as an all-purpose assistant.
Currently, the list of apps available to use through ChatGPT, on an invite-only basis, is small. Expedia, Instacart, Kayak, Klarna, OpenTable, Shopify, and Slack are probably the best known, and there are tools for accessing financial and legal data and doing mathematical calculations. The suggestion, however, is clear: ChatGPT is going to at least try to see if it can function as a new home page, or home screen, for its users — as an agent that mediates or automates regular computing tasks, as an interface for your other interfaces. Instead of going to a website, interacting with its interface, and finding or doing what you need, ChatGPT will attempt to carry out and conceal that process for you.
OpenAI isn’t the first company to try to get its users to use chat instead of buttons, search boxes, and menus, but previous attempts to retrofit popular U.S. services with conversational interfaces haven’t gone well. (Remember “chatting with the news“?) Granted, previous chatbots were comparatively primitive. And what OpenAI is testing out here isn’t just whether chatting is a better way to engage with, say, a job-listing platform but whether it can turn its users into captive customers who start and end their days in OpenAI’s text box instead of Google’s. It’s an attempt to manifest the Silicon Valley dream of an everything app: one piece of software through which you use all others, an operating system for everything, a fundamentally different way of interacting with our computers and phones, and a thing that replaces the home screen, search box, feed, or maybe all of the above.
This is a significant announcement for a few reasons. First, it suggests a plan for OpenAI that isn’t just licensing agreements with existing software companies or charging for access to an API. ChatGPT with an app store is a real product aimed directly at regular people, not a testing environment or tech demo. It gives a better sense of what OpenAI might actually try to do as a company than a hundred aimless conversations about AGI, x-risk, or machine consciousness. It feels like an actual encounter with OpenAI as the well-funded start-up it is rather than as a research lab, series of viral social-media phenomena, or avatar of budding new tech epistemology. It’s a plan to give people a different way to use tools they already need and like and to give other companies a way to build new things in an ecosystem. If OpenAI’s partnership with Microsoft shows what its AI looks like as a feature in other companies’ platforms, ChatGPT plug-ins are an attempt to become a platform.
This is, broadly speaking, how every previous internet giant became dominant: by giving outside firms a way to get a piece of their growth, giving users a way to do more from within their apps and services, and eventually accumulating and leveraging their new powers over firms that have become overly dependent on access to their platf — well, I’m sure it’ll be different this time around.
Maybe chat as a general computing interface will quickly discover the limits of its appeal, sending OpenAI’s GPT models, and billions of dollars, running in a different direction. Or we could find that giving executive powers to an early-stage AI that’s increasing in capability on an “exponential” curve proves to be as reckless as it sounds in summary (there are obvious risks here for plug-in creators, which will learn firsthand what exactly a hungry LLM can do with access to their services and users).
In any case, the ambition here is massive and clear: OpenAI is wondering aloud whether it might be possible to subordinate and remake the internet with an everything chat — and declaring its intention to try.