[exe34]

if you can do 100x more efficient training with open source, closeAI can simply take that and train a model that's 100x bigger/longer/more tokens.

[bugglebeetle]

AKA why Unsloth is now YC backed for their even better (but closed source) fine-tuning.