Extract from Cat Casey’s article “GPT-4o’s Flirty Fail: Why AI Shouldn’t Imitate ‘Her'”
Last week, millions of ChatGPT fans and I counted down the minutes for the long-awaited unveiling of OpenAI’s latest not-so-super-secret new release, GPT-4o (“o” for “omni”). The livestream started with a vision of omni-input, real-time, interactive generative artificial intelligence. GPT-4o touted a 232-millisecond response time, an eye-watering 50% cost reduction, functionality in many more languages, and was available to even free users and via API.
To say I was psyched is an understatement!
And then, at approximately the 10-minute mark, during the live demo of real-time multimodality interaction with the AI chatbot … I got the ick.
Something about the vocal fry, the cheeky pander, and the nervous laughter piqued a memory. Was this AI flirting with the demonstrator? And more pointedly, was the voice a near-deepfake-quality replica of Scarlett Johansson’s iconic portrayal of a sentient AI operating system-turned-paramour?
My suspicions were confirmed moments later by a three-letter X post from Sam Altman, the OpenAI CEO: “her.”
