to absolutely no surprise, the guy in charge of meta's bot rules, that allowed llms to have 'sensual' conversations with children, has been caught trying to abuse children.
This is exactly the same issue as doing blind translations of human languages; you may well get the idea across, but you're going to sound like an illiterate dipshit doing so, and there's going to be structural issues impeding your ability to get anything actually done with that.
As a blue teamer, this is a wholly unacceptable situation; "legal" is a risk surface, and now there is interference from a third-party service provider that substantively raises that risk profile's likelihood and severity.
Summary, an LLM bot nonconsensually generated replies to emails - making factual mistakes - in a context where there are legally binding consequences for statements made over email.
yeah, that more or less fits, especially how there's lots of folks using them because that's what work bought for a transport fleet, and a lot of folks using them because they're large and obvious social signals, and some assholes who have decided that 'rolling coal' is the way to go; and plenty of people who are taking them into city streets that were never designed for that kind of vehicle....