Anthropic chief says AI could surpass “almost all humans at almost everything” shortly after 2027

Which AI models, though? Your synthetic text extruder LLMs that can't accurately surpass humans at anything unless you train them specifically to do that and which are kinda shite even then unless you look at it exactly the right way? Or that fabled brain simulation AI that doesn't even exist?

Instead, he prefers to describe future AI systems as a "country of geniuses in a data center," [...] [and] that such systems would need to be "smarter than a Nobel Prize winner across most relevant fields."

Ah, "future" AI systems. As in the ones we haven't built yet, don't know how to build, and don't even know whether we can build them. But let's just feed more shit into Habsburg GPT in the meantime, maybe one will magically pop out.