[CrypticShift]

The trajectory we're seeing with quality small AI Models, coupled with the self-imposed censorship and the foreseeable scarcity of high-quality training data due to new copyright law, leads me to forecast a surge in "pirate" models.

Increasingly, the distinction between core model training and fine-tuning might become ambiguous (how ?). Considering this, we might witness a trend where custom 'add-ons' for AI models become commoditized. Imagine simply downloading a "New York Times" pack to enhance your unofficial "pirate" language model.