i think in the long term the problem is going to be this: a great small model always come from a greater large model, but the larger base model keep getting larger and more closed sourced

so long as there's no algorithm breakthrough