In addition to announcing a revision of the chat limits implemented on Bing’s ChatGPT today, Microsoft revealed how the company’s most recent Search Engine, new Bing, operates. The technology behind the new Bing is powered by the company’s new artificial intelligence (AI) model, dubbed Prometheus.
Microsoft announced Prometheus, a first-of-its-kind AI model that combines the company’s comprehensive Bing index, ranking, and answers results with the creative reasoning abilities of OpenAI’s most advanced GPT models, in a company-branded blog post published today.
In milliseconds, Prometheus uses Bing and GPT to generate a series of internal queries iteratively via a component called Bing Orchestrator, and then provides an accurate and rich response to the user’s query within the given conversation context.
Selecting the relevant internal queries and leveraging the respective Bing search results is a crucial aspect of Prometheus, as it provides the model with relevant and up-to-date information, allowing it to respond to recent questions and reducing inaccuracies. In other words, the model applies its reasoning ability to the Bing data and is therefore grounded by Bing data via the Bing Orchestrator. Microsoft has provided the diagram below to illustrate the overall operation of Prometheus.
Microsoft limits Bing chats to 5 questions per session
Prometheus attaches relevant Bing search results, such as weather, stock, sports, and news, to the Chat response as the final step. In consideration of the search results, Prometheus is able to integrate citations into the Chat response so that users can easily access and verify the information.
Microsoft’s Corporate Vice President of Search & AI, Jordi Ribas, stated in a blog post that, in addition to Prometheus, his team faced obstacles when developing the ideal Bing search product. Because some queries would perform better in the traditional or Search mode, while others would perform better in the conversational or Chat mode, the team ultimately created a new Bing UX with a unified Search and Chat in a single interface.
Selecting the relevant internal queries and leveraging the respective Bing search results is a crucial component of Prometheus, as it provides the model with relevant and up-to-date information, allowing it to answer recent questions and reducing inaccuracies – this is known as grounding. In other words, the model reasoned over the Bing data and is therefore grounded by Bing data via the Bing Orchestrator. The diagram below depicts Prometheus’ operation at a high level.
