OpenAI's groundbreaking AI chatbot, ChatGPT, has taken a giant leap forward by gaining the ability to browse the internet in specific cases.
Introducing plugins for ChatGPT, OpenAI has unlocked a new level of functionality for the bot, granting it access to an extensive range of third-party knowledge sources and databases, including the vast expanse of the web itself. Initially available in alpha to select developers and subscribers of the premium ChatGPT Plus plan, OpenAI plans to gradually expand access to a larger user base and eventually provide API access.
Among the impressive suite of plugins, OpenAI's first-party web-browsing plugin stands out as the most captivating. This plugin empowers ChatGPT to fetch information from various online sources, enabling it to answer a wide array of questions posed to it. Gone are the days when ChatGPT's knowledge was limited to dates, events, and people up until September 2021. Now, with the Bing search API as its resource, ChatGPT can seamlessly integrate content from the web into its responses, ensuring transparency by citing the sources it consults.
While a web-accessible chatbot may sound like a risky endeavor, OpenAI has taken the necessary precautions to mitigate potential issues, as their own research has shown. Previous experiments, such as WebGPT and Meta's BlenderBot 3.0, encountered challenges by quoting unreliable sources and venturing into conspiracy theories and offensive content. Nonetheless, OpenAI is fully aware of the dangers and has implemented safeguards, backed by internal and external red teams, to prevent undesirable behaviors like fraudulent activities, spam emails, and bypassing safety restrictions. However, only time will truly test the efficacy of these measures.
It's important to acknowledge that the live web is inherently less curated than a static training dataset, presenting both advantages and challenges. While search engines like Google and Bing employ safety mechanisms to minimize unreliable content, their algorithms can be manipulated, leading to potential biases. Additionally, these algorithms favor websites that embrace modern web technologies, potentially overlooking websites with high-quality content. Consequently, search engines possess significant influence over the data that informs web-connected language models, with Google sometimes prioritizing its own services over alternative, more socially driven sources.
OpenAI's decision to introduce plugins to ChatGPT's repertoire is a significant development. Previously confined to its training data, ChatGPT's capabilities have expanded immensely, potentially alleviating concerns about unlicensed work. Some experts have criticized OpenAI for leveraging the vast amount of data from public websites in ChatGPT's dataset. However, plugins address this issue by allowing companies to maintain complete control over their proprietary data.
OpenAI's commitment to open sourcing a "retrieval" plugin further demonstrates their dedication to collaboration and improvement. By enabling ChatGPT to access specific sections of documents through natural language queries, OpenAI is fostering the creation of new plugins and welcoming contributions from the community. With everyone's input, OpenAI aims to build a useful and safe environment.
The introduction of plugins marks a crucial milestone in ChatGPT's journey. From relying solely on its training data, ChatGPT has now become remarkably versatile and potentially less susceptible to legal concerns. While some raise questions about the licensing of data used for ChatGPT's training, plugins pave the way for data control and integrity, providing a solution that benefits all parties involved.
