OpenAI pauses ChatGPT-4o voice that fans said ripped off Scarlett Johansson
arstechnica.com
"Sky’s voice is not an imitation of Scarlett Johansson," OpenAI insists.
  • JesusEnglish
    30•7 months ago

    IMHO, the ScarJo voice was the least weird part. The weirdest part was the intense horny energy that they trained into the model. It’s like they trained it on the first 5 minutes of every girlfriend role play video on porn hub.

    • @lorkanoEnglish
      8•7 months ago

      I don’t agree she or any of the voices sounded horny to be honest. But they sounded overly movie / voice actor like / emotional. People do not speak like this in every day basis.

  • @phoneymouseEnglish
    22•7 months ago

    Knowing OpenAI, they probably did just rip off her voice.

    • @lorkanoEnglish
      1•7 months ago

      Knowing every single Ai company probably

  • FubarberryEnglish
    17•7 months ago

    I feel kinda bad for the voice actor they used. Her voice was (potentially) going to be famous as the “AI” voice (similar to Siri), and instead that won’t happen because she happens to sound like Scarlet Johanson.

    • @Num10ckEnglish
      4•7 months ago

      it would be a good PR move if they released a statement from the actress who trained it and had some agreement with them. otherwise i’ll continue to believe they owe Scarlett a royalty.

      • @lorkanoEnglish
        2•7 months ago

        Might be too late for it to make any meaningful change, now everyone already says it’s Scarlett

  • MentalEdgeEnglish
    8•7 months ago

    My initial reaction to the announcement clips was that of disgust.

    Like auditory uncanny valley. The result is impressive, but holy fuck hearing it speak makes me cringe, I want to shut it up asap.

    Unplug it before it can ask if I want to kill myself, as if it actually cares, and isn’t a million coin tosses away from acting on some random training data to encourage me to do it instead of not.

    To interact with humans with such sycophancy, I am viscerally creeped out imagining the design objectives.

    The worst part is, it’s being presented with the same false confidence that LLMs themselves exhibit… Even as they confidently explain how simple it is to prove the earth is flat when you ask.

    Fucking hell, the tech is amazing, but can we not handwave away the limitations and irresponsibly apply it to literally everything, as if just a few more yottabytes of training data means it’ll stop making mistakes “aaaany second now”.