Tomasz Tunguz

I actually disagree. I believe that we can get a better underlying theory of what is going on with LLMs on a fairly abstract level. The question is if that theory could help us at all with making LLMs more efficient.