product is largely hamstrung by the fact that it has to allow children, nuns, retired colonels, and US Christian right-wingers to use it. This means that many words in common use, which are ideal for describing situations, or scenes, are banned because what they might be used for. Yesterday DALL-E would not create a picture of the Ancient Greek goddess Persephone because I described her as wearing a black robe. Apparently robe was a blocked word.
Puritans involved in AI insist that is to prevent the creation of deep fakes, but others point out that taking care of conservative morals does fit with OpenAI's pledge to create a "safe and beneficial" AI.
OpenAI is toying with the idea of giving devs and users the green light to "responsibly" churn out what they're calling not-safe-for-work (NSFW) content with their products. They're talking about stuff like erotica, over-the-top gore, nasty language, and swearing.
It an announcement the company said that it was having a nosy into whether we can give the thumbs up to NSFW content for those old enough to see it...
“We're keen to get a handle on what people and society reckon about how our models should behave when it comes to this sort of thing."
The comment was part of a blurb OpenAI put out about how they put their AI tools together. Joanne Jang, one of the OpenAI lot in San Francisco who had a hand in the write-up told NPR OpenAI wants to spark a conversation about whether it's right always to keep erotic text and birthday suit snaps off its products. She made it clear that deepfakes are still a no-go.