The crux of the problem is that in a few years, if LLM capabilities continue to increase at the current trajectory, it actually might go great