Web search with RAG // SearchGPT
Why not re-index all web information into a vector database and perform a similarity search and LLM-generation // real-time
SearchGPT is a prototype search feature combining AI capabilities with real-time web information to provide quick, relevant answers with clear sources. Launched for a limited group of users and publishers to gather feedback, the aim is to integrate its best features into ChatGPT. This tool is designed to simplify web searches, provide direct answers with links to sources, and allow follow-up questions in a conversational manner.
SearchGPT partners with publishers to enhance content discovery and engagement while protecting journalism. The prototype includes tools for publishers to manage their appearance in SearchGPT and offers insights into AI search interactions. Feedback from this prototype will inform future improvements and integration into ChatGPT. Users can join the waitlist to try SearchGPT.
Bing introduced LLM-powered chat answers in February last year and is now evolving search with a new generative search experience, currently available to a small percentage of queries. This approach combines generative AI and large language models (LLMs) to create dynamic, AI-generated responses alongside traditional search results. For instance, a search for “What is a spaghetti western?” produces detailed, easy-to-read information on the film subgenre with links and sources.
The generative search experience integrates Bing’s search foundation with LLMs and small language models (SLMs), analyzing millions of sources to generate accurate and relevant search results. Early data shows that this new feature maintains website traffic and supports a healthy web ecosystem by including more clickable links.
Bing is carefully rolling out this feature, seeking user feedback through thumbs up/down icons and a Feedback option, and will continue refining the experience before wider release. More updates will be shared in the future.
