Comment by zbshqoa

Number 2 is already possible with open models. You can do distillation using Llama, which could likely be doing #1 to build their models (I'm not sure it's the case though)