Why? For Super AGIs with continuous ML/inference (aka artificial brain)?
As far as I know those models do not exist yet. Only models with a split between ML and inference (going to be really hard to get weight stability without a "sleep" regime).
OpenAI is betting on scaling, which means using more compute per request. More users times more compute is a lot more compute needed. We don't need to assume specifically what techniques will be used.
5 comments
[ 2.4 ms ] story [ 18.5 ms ] threadAs far as I know those models do not exist yet. Only models with a split between ML and inference (going to be really hard to get weight stability without a "sleep" regime).