I knew that part and assumed they were training on repos, but I didn’t know they linked them so directly internally. Like this is the most important asset. Train on every push kinda thing.
I was expecting that as well but somehow Microsoft botched it. Training on failed commits should be the most invaluable asset. Yet their internally developed models (Phi) are lagging back of the competitors.
I knew that part and assumed they were training on repos, but I didn’t know they linked them so directly internally. Like this is the most important asset. Train on every push kinda thing.
I was expecting that as well but somehow Microsoft botched it. Training on failed commits should be the most invaluable asset. Yet their internally developed models (Phi) are lagging back of the competitors.