|
|
|
|
|
by meander_water
55 days ago
|
|
I think most labs actively create synthetic data using existing model as part of the mix for the pretraining stage for their next model. Would love to know exactly what the latest process is to keep slop out of training data. |
|
:)