LESSWRONG
LW

All of Wis3man's Comments + Replies

A breakdown of AI capability levels focused on AI R&D labor acceleration
Wis3man110

I think the compute bottleneck tax is probably around 25% with this acceleration (really, we should think about this in terms of tax brackets).


Please could you expand on why we should think in terms of tax brackets? 

Reply
6ryan_greenblatt
I expect that as you increase AI R&D labor acceleration compute becomes a larger and larger bottleneck. So, the first doubling of acceleration has less of a compute bottleneck tax than the 4th doubling. This can be equivalently thought of in terms of "tax brackets" though perhaps this was a confusing way to put it.