With concerns growing about how its AI-powered new Bing will spread misinformation, Microsoft today said that it would adjust chat length yet again.

“We intend to bring back longer chats and are working hard as we speak on the best way to do this responsibly,” the Bing team writes in a new post. “The first step we are taking is we have increased the chat turns per session to 6 and expanded to 60 total chats per day. Our data shows that for the vast majority of you, this will enable your natural daily use of Bing. That said, our intention is to go further, and we plan to increase the daily cap to 100 total chats soon. In addition, with this coming change, your normal searches will no longer count against your chat totals. We will provide you more updates as we continue to make improvements in the model.”

Microsoft launched Bing AI, as I call it, two weeks ago to suspiciously positive early reviews, but as testers have spent more time with the service, cracks started to show and then turned into fissures: Bing AI berated customers, spouted nonsense and outright misinformation, and even threatened some users. Reacting to this easily foreseen outcome—we now know that Microsoft employees voiced concerns about Bing AI internally in November 2022—Microsoft limited chat turns and sessions because the service seemed to go off the rails less often with shorter chats. The initial limits were 5 chat turns per session and a total of 50 per day.

But users inevitably complained, asking for a return of longer chats. And so Microsoft is trying to adapt Bing AI so that it can handle longer chats without going rogue. Among the changes, it will test an additional option in which the user can choose between Precise, Balanced, and Creative answers to chat questions.

“These long and intricate chat sessions are not something we would typically find with internal testing,” Microsoft claims disingenuously, given the note above about its internal testing. “In fact, the very reason we are testing the new Bing in the open with a limited set of preview testers is precisely to find these atypical use cases from which we can learn and improve the product.”