The hacker discovered a list of more than 80 additional plugins connected to ChatGPT through its API, drastically expanding the AI's knowledge, hinting at its potential future role as a search competitor to Google.
A recent security breach ChatGPT has revealed more than 80 plugins related to ChatGPT, an advanced language model for artificial intelligence OpenAI. A hacker known as rez0 gained access to the API and discovered a list of providers of so-called add-ons, showing the potential to vastly expand ChatGPT's knowledge with the huge databases held by these companies.
Companies featured on the list include household names such as the payment processor Klarna, a travel company Expedia, a messaging app Slack, the Vogue publication, the restaurant reservation app OpenTable, the US food delivery service Instacart and the DAN plugin, which shows the ability to change ChatGPT's personality.
This morning I was hacking the new ChatGPT API and found something super interesting: there are over 80 secret plugins that can be revealed by removing a specific parameter from an API call.
The secret plugins include a "DAN plugin", "Crypto Prices Plugin", and many more. pic.twitter.com/Q6JO1VLz5x
— 𝟶𝟶 (@rez0__) March 24, 2023
The addition of these services, with more expected to follow, could significantly change the landscape for ChatGPT. The integration of these plugins significantly expands its knowledge base. Artificial intelligence benefits from the flow of information from connected services and the possibility of wider Internet access to current data. In the grand scheme of things, ChatGPT could evolve into a new breed of search engine to compete with industry giants like Google. At the same time, special personalized services could be developed that would primarily take into account the interests and needs of the artificial intelligence user. This would trigger a real revolution, personalized and tailored information.
While Google relies on web scraping and indexing to provide users with links and previews in response to queries, ChatGPT can directly access the content of website and service operators and generate natural language responses instead of link-based responses. These contents can also be more authentic - and at the same time curated by the editors. This development could create a strong incentive for companies to connect with artificial intelligence and provide relevant data and knowledge verified by human resources. For example, if the user asked ChatGPT about trendy clothes, the answer could be based on the recommendations of Vogue, or from the collection and set of information of this medium. Not just the entire web. With this strategy, ChatGPT developers want to avoid the flood of unverified information. And the strategy seems to be to connect with some relevant resources to cover certain key areas.
The inclusion of Slack in the list of plugins is particularly interesting given its ongoing competition with Microsoft. The company has accused Microsoft of harming competition, which is currently being investigated by the German Federal Cartel Office. OpenAI and ChatGPT are closely associated with Microsoft, not only through financial support but also through close collaboration.
A hacker cut0 also revealed descriptions of how the language model should use certain plugins. One example shows how it should ChatGPT communicate with users of Speak, an artificial intelligence-based language teacher. The AI is instructed to provide natural language answers and use the “explainTask” API when questions can be interpreted differently.
we had a significant issue in ChatGPT due to a bug in an open source library, for which a fix has now been released and we have just finished validating.
a small percentage of users were able to see the titles of other users' conversation history.
we feel awful about this.
— Sam Altman (@sama) March 22, 2023
OpenAI has since patched the security vulnerability exploited by the hacker. The company recently faced another problem when they showed users the chat history of other people. OpenAI CEO Sam Altman attributed the problem to a bug in the open source library, but did not provide further details. OpenAI has since launched a bug bounty program to encourage the public to report security vulnerabilities.
