Will AI training bifurcate into pre-training and prompt engineering? or will there continue to be one level (or more) of fine-tuning in between?