Date With Disaster There’s heartbreak in the buzzing world of AI, because legions of lonely OpenAI fans have just been stood up by, well, “Her.” The company first teased the expressive — and flirty? — Voice Mode assistant for ChatGPT, which could engage in bubbly, real-time conversations and even read facial expressions, back in May. At the time, the feature was expected to roll out in an alpha test to a select group of paid ChatGPT Plus subscribers by June. But now, in an announcement on X-formerly-Twitter, OpenAI says its Voice Mode needs one more month “to reach our bar to launch,” with a wider launch to all Plus subscribers in the fall — a move that’s led to even more controversy around the AI assistant. Saying No Some of the stated reasons for the delay are boilerplate. OpenAI said, in vague terms, that it’s “improving the user experience,” as well as readying its infrastructure to handle the incoming millions of users who will eventually engage with Voice Mode. Another reason that it snuck in, though, could raise eyebrows: OpenAI is also, cryptically, “improving the model’s ability to detect and refuse certain content.” That sounds like the Voice Mode is prone to some bad behavior and could constitute a far deeper safety — or “alignment,” in the AI sector’s euphemistic parlance — problem that has plagued chatbots in general, where they spout inappropriate or dangerous content either unprompted or at the deliberate request of a user. (We won’t rule out people simply being…OpenAI Fans Slam Delay of ChatGPT Voice Mode, as Company Hints at Trouble Behind the Scenes