On Thursday, OpenAI announced the launch of ChatGPT’s new plugins, which allow the chatbot to integrate with third-party APIs and enhance its abilities to respond to various situations defined by developers. These plugins follow the release of the new GPT-4 engine and Whisper API in March, and expand ChatGPT’s range of actions.
Prior to the latest GPT-4 upgrade, developing a sports chatbot would have limited the bot’s ability to discuss only past games and scores from 2021, as that was the extent of GPT-3’s training data. The chatbot would have been unaware of current real-time data or even the existence of the year 2022. However, with a ChatGPT plugin, developers can integrate ChatGPT’s functionality into their existing code stack, enabling the bot to perform tasks such as retrieving real-time information like sports scores, stock prices, and breaking news, as well as accessing specific knowledge-base information such as internal company documents or personal cloud data. The chatbot can even execute actions on behalf of the user, such as booking a flight or ordering take-out, similar to an installable Google Assistant created by OpenAI.
“The AI model acts as an intelligent API caller. Given an API spec and a natural-language description of when to use the API, the model proactively calls the API to perform actions,” the OpenAI team wrote. “For instance, if a user asks, ‘Where should I stay in Paris for a couple nights?’, the model may choose to call a hotel reservation plugin API, receive the API response, and generate a user-facing answer combining the API data and its natural language capabilities.”
OpenAI has highlighted that the use of plugins can help reduce the tendency for AI models to hallucinate facts while answering complex questions, by bridging the knowledge gap between what the model was trained on and what has occurred since. The company also stated that these references not only improve the model’s usefulness, but also allow users to evaluate the reliability of the model’s output and verify its accuracy, potentially reducing the risks associated with excessive dependence on the model. The team emphasized the importance of trustworthiness in the output of AI models, which can be addressed through the use of plugins.
The expanded capabilities and knowledge gained through the use of ChatGPT’s plugins also raise the possibility of the model generating problematic responses. To prevent the kind of financial loss that Google incurred with Bard, OpenAI has reportedly conducted extensive stress tests on these plugins. The team has performed internal and external red-teaming exercises that have identified potential areas of concern. They intend to use these findings to develop safety-by-design measures that improve transparency and limit the plugins’ ability to engage in risky behaviors.
The plug-in itself is still in early alpha with limited availability. OpenAI granted early access to a handful of partner companies including Expedia, Instacart, KAYAK, OpenTable, Shopify, Slack, Wolfram, and Zapier for use in their existing apps. You’ll need to add your name to the waitlist to try it for yourself.
