Google DeepMind's David Pfau argues that tech giants' commitment to unlimited AI spending guarantees a financial bubble
Prakash predicts a major architectural breakthrough within 18 months.
I don't know if the amount of money *currently* being invested is justified or not. If it turns out it is justified, people will double down and spend more. Constant double-or-nothing bets cannot pay out indefinitely.
The second people stop thinking it's a bubble, they invest even more. Almost every major tech company has said they will spend on AI until they literally can't spend any more. It's guaranteed that this will scale until it is, in fact, a bubble.
This doesn't mean I think AI capabilities will stop improving if you continue scaling. I think at a certain point the economic value derived from the increased capabilities will not outweigh the increased cost of scaling.
I don't know if the amount of money *currently* being invested is justified or not. If it turns out it is justified, people will double down and spend more. Constant double-or-nothing bets cannot pay out indefinitely.
I don't know when that point will be.
This doesn't mean I think AI capabilities will stop improving if you continue scaling. I think at a certain point the economic value derived from the increased capabilities will not outweigh the increased cost of scaling.
I think what will likely happen is a breakthrough as big as the transformer within 18 months and potentially within 6. A sudden step up in capabilities using current compute, and a reenergized bull market. Things are not going be linear from here on out.
The second people stop thinking it's a bubble, they invest even more. Almost every major tech company has said they will spend on AI until they literally can't spend any more. It's guaranteed that this will scale until it is, in fact, a bubble.