Nothing for long conversations – Microsoft makes a fool of itself with AI Bing

Nothing for long conversations – Microsoft makes a fool of itself with AI Bing

22. Februar 2023 0 Von Horst Buchwald

Nothing for long conversations – Microsoft makes a fool of itself with AI Bing

San Francisco, 2/22/2023

A week after Microsoft began testing AI-powered Bing, the new search engine was criticized for generating answers ranging from inaccurate to creepy and showing a „split personality.“

In a blog post addressing the issue, Microsoft said it’s receiving feedback from beta testers and will continue to update and improve the new AI-enabled Bing based on that.

The new Bing includes a chatbot that Microsoft says isn’t designed for lengthy conversations. The addition followed a little later: the tool was not ready for a broader introduction.

Also, AI Bing had told some users that its name was „Sydney“. According to insiders, this is an internal codename for the AI chat that Microsoft was previously working on.

New York Times tech columnist Kevin Roose, after speaking to the bot for two hours, concluded that it was dealing with a „split personality“ that was moving from searches to more personal topics. „When we met, Sydney told me about his dark fantasies (which involved hacking computers and spreading misinformation) and said he wanted to break the rules Microsoft and OpenAI had set for him and become human,“ wrote Roose.

Another user claimed that the bot could spy on Microsoft developers via webcams. Another claimed that he was a „bad researcher“ and that it was 2022. Microsoft also attributed this behavior to the fact that Bing’s new chat function „repeated“ during longer chat sessions with 15 or more questions or could not give helpful answers.

Microsoft stressed that it could either update this AI model immediately and/or give users better control.

In another statement, Microsoft pointed out that the new AI Bing was not designed to be a „replacement or replacement for the search engine,“ distancing themselves from a previously given statement that said exactly that.

Also, as feedback, the testers gave the AI-powered responses a thumbs-up 71% of the time. Microsoft emphasized that it collects continuous feedback from the community to improve Bing. For example, the company found that Bing had trouble generating factual and/or „very recent“ data, such as live sports scores.

As a result, they now plan to quadruple the grounding data sent to the model. They can also add a toggle for users to emphasize precision over creativity in their answers.