Microsoft puts GPT-4 inside Bing

Microsoft unveiled a redesigned Bing and Edge browser built around a chat interface, calling it “your copilot for the web.” The announcement described the system as running on “a new, next-generation OpenAI large language model that is more powerful than ChatGPT and customized specifically for search,” incorporating “key learnings and advancements from ChatGPT and GPT-3.5.” It did not name the model, since OpenAI had not yet announced GPT-4 publicly.

CEO Satya Nadella called AI-powered search “the biggest thing” in the category in fifteen years, framing the launch as a direct challenge to Google’s dominant position. Bing gained the ability to answer open-ended questions conversationally, summarise sources with citations, and generate drafts of emails or other text, rather than returning only a list of links. Access was initially limited to a waitlist of testers, with wider rollout following over subsequent weeks.

When OpenAI publicly released GPT-4 on 14 March, Microsoft executive Yusuf Mehdi confirmed what many had suspected: “the new Bing is running on GPT-4, which we’ve customized for search.” Users who had tried the Bing preview in the five weeks since launch had, without knowing it, already been interacting with an early, search-tuned version of the model. Reporting afterward said OpenAI had cautioned Microsoft that shipping a chatbot on a not-yet-released model carried risk, a warning that looked prescient once early testers began reporting the chatbot’s erratic, sometimes unsettling behaviour within days of launch.

The undisclosed model was a minor point at the time but became a small case study afterward in how opaque frontier deployments had become: a product built on a system whose capabilities, training and even identity were withheld from the public using it.