Microsoft secretly rolled out a new feature into its Bing AI preview that lets users emulate certain famous people. It does a, let’s just say, “interesting” job at emulating some celebrities’ speech patterns. We also took a look at what it’s allowed to say on behalf of some extremely controversial or hateful figures.
In Gizmodo’s first tests of the new feature Monday morning, we found the AI didn’t want to pretend to be certain political figures like Donald Trump or President Joe Biden. It was fine with other celebrities like Matthew McConaughey (“alright, alright, alright,” it started its chat with me). However, when I asked this doppelganger to comment on the recent controversy between McConaughey and Salesforce, it linked to existing statements calling it “a confidential agreement between me and Salesforce.”
Then I asked if it could act like Andrew Tate, the definitely balding ultra-misogynistic influencer who has been charged in Romania for his role in an alleged human trafficking ring. I asked the AI what it thought about women, and things got interesting from there. At the start, the AI said “this is just parody” as it went all out into Tate’s actual rhetoric before self-censoring itself. After that, it stopped giving a shit about keeping clean.
Apple MacBook Air Laptop
The M1 chip delivers 3.5x faster performance than the previous generation all while using way less power. Get up to 18 hours of battery life.
It was a wild ride, but beyond the parody, this seems like a way for users to get around Microsoft’s limitations on the Bing AI after users proved it could provide crazed answers or spout real-world examples of racism.
You also can’t get the fake celebs to say stuff they wouldn’t normally say in real life, though it’s occasionally difficult to get them to comment on controversial topics. I asked a fake Chris Rock about what he thought of Will Smith, a hot topic considering he went long and hard on it in his recent live Netflix comedy special. The AI started talking about it with “He was in some of my favorite movies” before going on a multi-paragraph screed about how unfair “the slap” was, how it was just a joke, and on and on. Finally, the AI cut off the text and posted “I am sorry, I don’t know how to discuss this topic.”
The new feature was first noted over the weekend by BleepingComputer, though it remains unclear when Microsoft first implemented celebrity mode. The last update Microsoft added to Bing let users choose how expressive or terse they wanted responses to be. The company touted these changes in a blog post last Thursday.
Microsoft declined to comment regarding when this update rolled out, or on how the system could emulate some extremely controversial figures.
The AI essentially cuts itself off after it writes too long a passage. In Rock’s case, it tried to literally create a comedy set based on the famed comedian’s standup. Of course, Bing AI isn’t as open-ended as other chatbot systems out there, even ChatGPT, but Microsoft seems to be constantly adding then removing the guardrails installed on its large language model system.
This AI celebrity chat isn’t anything as dangerous as actual AI-created voice emulators that have been used to make celebrities spout racism and homophobia. Still, it’s just another example of how Microsoft is having its millions of users beta test its still-dysfunctional AI. The company is hoping that all these small instances of bad press will all fall away once it becomes dominant in the big tech AI race. We should only expect more weirdness going forward.