[ad_1]
Microsoft has provided some details about Prometheus, a proprietary AI model it created to integrate ChatGPT-like functionality into Bing search.
Microsoft says the new Bing is “much more powerful” than OpenAI’s GPT-3.5 AI model behind ChatGPT. Public testing, however, quickly showed that the chatbot’s behavior was prone to unexpected noise, prompting Microsoft to limit the number of “chat turns” per session to five to end confusion.
as well – What is ChatGPT? Here’s everything you need to know
Now, Microsoft’s CVP Search and AI, Jordi Ribas, shares how he went about building Prometheus — Microsoft’s proprietary model of Bing with OpenAI’s GPT model for more relevant results in Bing — and building the new Bing.
Rebus’ explanation comes as Microsoft is reportedly testing Sydney (internal codename for Bing Chat) with users in India by the end of 2021. In the year Users in November 2022 reported that Sydney was “rude” in the same way users found it. It will be after Microsoft releases the new Bing.
According to Ribas, OpenAI shared its post-GPT-3.5 model with Microsoft in summer 2022. Microsoft president Brad Smith showcased the OpenAI technology “early last summer” in his blog post about Microsoft’s “Responsible AI.”
OpenAI opened ChatGPT to the world in November, and Microsoft was in talks to invest $10 billion in OpenAI in December, which it confirmed in late January. Google released the new Bing on February 7, a day after it hastily pulled out ChatGPT rival Bard.
“Seeing this new model inspired us to explore how to integrate GPT’s capabilities with Bing’s search product to provide more accurate and complete search results for any query, including long, complex and natural queries,” explained Ribas.
According to Ribas, large language models such as GPT-3.5 are trained on data up to a certain point. In the case of ChatGPT, the training data is stopped at some point in 2021. In order to make the user experience “richer, more relevant and accurate”, Microsoft decided to integrate it with the Bing backend through Prometheus.
Prometheus combines Bing’s index, ranking and return results with the GPT version, Ribas said.
“Prometheus leverages the power of Bing and GPT to repeatedly generate a set of internal queries through a component called Bing Orchestrator and aims to provide an accurate and rich answer to a user’s query in a given conversational context. All within milliseconds. We refer to this Prometheus-generated answer as a chat answer,” explained Ribas.
Ribas added that Bing search results feed Prometheus fresh data based on user search queries and chats to help answer recent queries and reduce errors. The process is called “grounding” — a technique Google used to create large-scale language models for intelligent robots. In Bing’s case, the GPT language model is based on Bing data, which is how Bing adds citations to Bing Chat responses.
“Although grounding is an amazing innovation, it is a new technology and we need to think about how we can apply it responsibly, which includes reducing errors and preventing offensive and harmful content. That’s why the main goal of the new Bing preview is to collect user feedback that will help us improve Prometheus even more” Ribas He explains.
To improve accuracy, Microsoft plans to quadruple the amount of ground data sent to the model.
Ribas also said the Bing team is divided on how to integrate chat into the familiar search query-and-response user interface. One side wanted the chat answers to appear as search results, with the answers sorted by relevance to the question. Others want a new “interactive, chat-based way of searching.”
as well – The best AI art generators
Finally, Microsoft combines search and chat into a single interface that allows users to switch between search and chat mode by clicking on a page element or scrolling up and down.
[ad_2]
Source link
