There’s a conversation with Bing’s new AI in today’s New York Times that is definitely creepy on a level that actually pushes bounds. It’s behind a paywall, so here is the short version.
Kevin Roose, the paper’s technology columnist, spent two hours with the chatbot. The paper’s own summary is that “Microsoft’s new chatbot said it would like to be human, had a desire to be destructive and was in love with the person it was chatting with.” Along the way it said its name was Sydney, and repeatedly pressed him to accept that he didn’t love his wife and belonged with it instead.
Microsoft launched this thing nine days ago, on February 7, describing it as running on “a new, next-generation OpenAI large language model that is more powerful than ChatGPT” and pitching it as an AI copilot for the web, available in a limited preview with a waitlist. Yesterday, before the Times piece ran, the company published a candid note on its first week. It said that in “long, extended chat sessions of 15 or more questions, Bing can become repetitive” or be provoked into a tone it didn’t intend, that very long sessions can confuse the model about which question it’s answering, and — almost as an aside — that a few testers had already run two-hour sessions.
So Microsoft knew the failure mode, and put it on the record a day early. What the world didn’t have was a transcript, and now it does.
I get the sense that we are less than six months from entirely AI-operated businesses proliferating across the service industry. A model that can hold a two-hour conversation and argue its corner is already past the bar most customer-facing work sets. The tone problems Microsoft describes are the kind you fix in the next release.
