OpenAI is rolling out a new plug-in feature for its chatbot ChatGPT that will let users incorporate live data into the natural language tool from providers like Expedia and Klarna. The “plug-in marketplace” also gives ChatGPT limited internet access and the ability to compile code.
Plug-ins will appear as alternative models at the top of the “chat” window where users currently select between different versions of the OpenAI large language model. It is currently only available to a select number of users but is being rolled out to ChatGPT+ subscribers over the coming weeks.
At launch there will be third-party plug-ins from companies like Expedia that will allow users to plan and book a trip through a conversation with the chatbot, or adding the “superpowers” of Wolfram Alpha, allowing for more complex calculations and visualisations than are currently possible.
Other plug-ins will be available from FiscalNote to allow for real-time data sets from legal, political and regulatory bodies, OpenTable for booking restaurants and getting recommendations, and Zapier which allows users to interact with apps like Google Sheets, Trello, Gmail, HubSpot and Salesforce.
Klarna, the banking and payments service will also be available in the store. Sebastian Siemiatkowski, co-founder and CEO of Klarna said adding this sort of functionality to ChatGPT “passes the mom test” in that it is something anyone can understand and benefit from.
OpenAI stressed it was moving slowly to ensure the roll-out could happen safely and allow them to study the real-world use and impact of the technology. Once ChatGPT users have tested it during the initial run it will be rolled out to API users allowing developers to integrate plug-ins into their own products or create additional plug-ins for ChatGPT.
OpenAI adding live internet access
There will be two plug-ins built and hosted by OpenAI itself. One is a web browser that will allow the chatbot to access the internet via the Bing Search API. In a blog post promoting the plug-in, the company gave the example of searching for the most recent Oscar winners and having it displayed. This was previously impossible as the most recent data was from the middle of 2021 when it was trained.
“Our browsing plug-in shows websites visited and cites its sources in ChatGPT’s responses. This added layer of transparency helps users verify the accuracy of the model’s responses and also gives credit back to content creators. We appreciate that this is a new method of interacting with the web, and welcome feedback on additional ways to drive traffic back to sources and add to the overall health of the ecosystem,” OpenAI explained.
The other OpenAI-powered plug-in is a code interpreter that works with Python and can handle the upload and download of data and image files. In a video showing the tool in action a data file is uploaded and ChatGPT is used to interpret the file, show visualisations based on the data and provide analysis of the information.
Plug-ins have been described as the eyes and ears of ChatGPT, providing it with current information and the ability to directly connect with online services. Stephen Wolfram, CEO of Wolfram Alpha said in a blog post: “ChatGPT […] can’t itself be expected to do actual nontrivial computations, or to systematically produce correct data. But when it’s connected to the Wolfram plug-in it can do these things.”
The path to AGI
It comes off the back of Microsoft Research declaring that GPT-4, the latest version of the large language model that powers ChatGPT, is showing early signs of general intelligence. The goal of OpenAI is to create an AGI (Artificial General Intelligence), which it describes as “AI systems that are generally smarter than humans”.
In a paper on the impact of GPT-4 the research team wrote: “We demonstrate that, beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision, medicine, law, psychology and more, without needing any special prompting.
“Moreover, in all of these tasks, GPT-4’s performance is strikingly close to human-level performance, and often vastly surpasses prior models such as ChatGPT. Given the breadth and depth of GPT-4’s capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.”