[nwoli]

Scaling linear algebra in the end is probably all we’ll need in the end. Only missing data and compute to get there

[imtringued]

Memory capacity is a much bigger problem. Mixtral 8x22B is a 200GB+ model.