Recently, Google officially launched its new Gemini API URL Context feature, aiming to enable AI to have human-like understanding when browsing web pages. This feature was launched on Google AI Studio on May 28th, marking another major breakthrough in Google's AI technology field.
Different from the common link sharing, the operation of the URL Context feature is completely different. Usually, when we give a link to AI, it often only reads the content of the web page through simple browser tools or search engine plugins, and this method often only gets the summary or partial information of the web page. However, Gemini's URL Context is an API specifically designed for developers, which can accurately parse and understand all the content of a web page, and even process multiple formats such as PDFs and images.
Specifically, the URL Context feature can handle up to 34MB of web content and supports various file formats, including HTML, JSON, and CSV. Logan Kilpatrick, a product manager at Google, stated that this feature greatly simplifies the development workflow, allowing developers to achieve deep information extraction and processing with just a few lines of code. Compared to the traditional "retrieval - augmented generation" (RAG) process, URL Context reduces the cumbersome steps, no longer requiring additional content extraction, vector storage, and retrieval mechanisms.
This feature is very powerful. With a simple URL, Gemini can extract key data from Tesla's financial reports, such as "total assets" and "total liabilities," which are often difficult to obtain through traditional methods. In addition, it can identify complex structures in PDFs, including tables and footnotes. This allows developers to quickly and accurately obtain the required information, greatly improving work efficiency.
However, URL Context also has its limitations. It cannot break through paywalls and is powerless against content that requires login to access. Additionally, it does not process specialized tools such as YouTube videos and Google Docs. The cost of content processing is billed based on the number of tokens, so developers need to reasonably design the information sources to control costs.
The URL Context feature not only demonstrates the cutting-edge development trend of AI technology but also gives us a new understanding of future information retrieval methods.
Article address: https://towardsdatascience.com/googles-url-context-grounding-another-nail-in-rags-coffin/
