@ToKTeacher@ChrisvanMerwijk I'm struggling to understand why universality implies a generally intelligent machine won't be dangerous? Does it matter if we can come up with the same explanations eventually when they can do it much faster in practice?
@badblankname@mitchellluns @lucysbeehive @esjesjesj and this is also why they run such "tight ships" there isn't room to make much more profit as its such a competitive industry
@badblankname@mitchellluns @lucysbeehive @esjesjesj All it really takes is one or two airlines to buck the trend and steal a whole lot of customers. It's the same reason why an airline won't just raise prices arbitrarily to "increase profit margins". There are a lot of other airlines waiting to steal your customers
@nickcammarata How are you finding hallucination rates? Doesn't a lot of the supposed convenience get wiped if you have to end up fact checking everything yourself anyway
@dwarkesh_sp@tylercowen I wish he went in a little deeper into this bottle neck argument. Like I'm still struggling to imagine the concrete bottlenecks that exist to keep growth down <0.5% when you have asi or whatever
@fchollet it seems to be that its entirely possible we can 'grow' AGI without understanding it enough to 'build' it ourselves. Like even now we kinda just let deep learning rip and it kinda just works
@fchollet Are we even able to understand the 'intelligence' of current LLMs? Like it seems they have non zero intelligence and we don't really have a great idea of what exactly is going on.