It was a marketing campaign. “We’re super cool. We’re super advanced. We hope don’t make the terminator! wink wink” “Oh no! GPT-2 is too dangerous! We better not release it! Oh wait. We did. ¯\_(ツ)_/¯ “)
It was never a serious thing. Sure here and in the depths of weirdo forums people whispered about the AGI that can’t be named — Roko’s Basilisk. It was just comically stupid. Pascal’s Wager with extra steps.
Now OpenAI has tons of money and everyone wants to deal with them, so they don’t need that hype machine. Instead they’re saying stick their “dangerous” models in your toaster Maximum Overdrive style.
It was transparent