[napier]

very different initial approaches to training dataset curation, to SD's detriment.

[GaggiX]

Can you be more explicit about what you mean? Also, the biggest difference is probably just the fact that MJ is using a much larger model than SD.