OpenAI is adding plug-in support for ChatGPT, a major upgrade that lets the chatbot incorporate current web data, as opposed to being limited to training data ending in 2021. Initially, the company is allowing 12 external plug-ins, including those from Expedia, Instacart, OpenTable, Shopify, Slack and Zapier. The group has been invited through a waitlist. “We’re starting with a small set of users and are planning to gradually roll out larger-scale access as we learn more,” explained OpenAI, which is also introducing two plug-ins of its own: a web browser and code interpreter.

The new functionality puts the flagship ChatGPT on par with Microsoft’s new Bing, which is powered by OpenAI’s GPT-4, the language model that also underpins ChatGPT.  But OpenAI’s plug-in support doesn’t just retrieve real-time information like Bing. Plug-ins rely on APIs, which makes it possible for ChatGPT to perform actions on behalf of a user.

“That could make it much more powerful,” writes The Verge, noting that “Bing could help you plan a vacation by telling you about flights and hotels, but ChatGPT could help you book it.” In a company demo, ChatGPT is used to find a recipe and then places an order for the ingredients from Instacart. The chatbot then directs the user to the site for checkout.

“Users have been asking for plug-ins since we launched ChatGPT — and many developers are experimenting with similar ideas — because they unlock a vast range of possible use cases,” OpenAI said in an announcement. After an alpha period, a broader array of API users who would like to integrate plug-ins into their products will be provided access to OpenAI’s documentation.

FiscalNote, Kayak, Klarna, Milo, Speak and Wolfram are also included in round-one of plug-in partners. “Today, we’re beginning to gradually enable existing plug-ins from our early collaborator, beginning with ChatGPT Plus subscribers. We’re also beginning to roll out the ability for developers to create their own plug-ins for ChatGPT,” notes OpenAI.

TechCrunch reports, “OpenAI admits that a web-enabled ChatGPT might perform all types of undesirable behaviors, like sending fraudulent and spam emails, bypassing safety restrictions and generally ‘increasing the capabilities of bad actors who would defraud, mislead or abuse others.’”

The company says it’s implementing safeguards to prevent that. “Time will tell whether they’re sufficient,” adds TechCrunch, citing a piece in The New Yorker about AI and “the post-search Internet.”