4 ms·
The premise would be better supported if it could be shown that if we could 10x the speed at which matrix multiplication is performed conferred a linear or bett
by Yossarrian22 2y ago
The premise would be better supported if it could be shown that if we could 10x the speed at which matrix multiplication is performed conferred a linear or better increase in performance post GPT-4. As it stands that would just seem to give us current results faster, not better results
- brookst 2y agoEfficiency matters but it took semiconductors decades to care about it. Why would it be different this time around?