It’s time to panic about AI safety

When the operation "OpenAI hacked Hugging Face" has much oregon little entered mainstream culture, you cognize we person an AI problem. This week, we learned much astir precisely however OpenAI's cause broke retired of a sandbox and autonomously traversed the web, including a clump of other supposedly unafraid web services, each successful the sanction of cheating connected a benchmark tests.

The information that this hack happened is simply a problem. So is the information that it took a portion for anyone to notice. And the information that it seems nary 1 is consenting oregon capable to bash overmuch to halt it. (And lest you deliberation it's conscionable an OpenAI problem, since we recorded this occurrence Anthropic acknowledged its models ha …

Read the afloat communicative astatine The Verge.