Mar 22, 2023·edited Mar 22, 2023Liked by Trevor Klee
This is a profound point. It really lets us understand that using LLMs as a "subroutine" within AI systems will be quite different than using any and all logic based systems as we have used them in the past.
We correctly understood these systems were "brittle" in ways that human reasoning was not.
Finally we have AI systems that are not brittle.... God help us.
... a "small" exercise remaining is to graft such an LLM substrate under the well understood brittle AI reasoning systems we have built over the last 70 years.
This is a profound point. It really lets us understand that using LLMs as a "subroutine" within AI systems will be quite different than using any and all logic based systems as we have used them in the past.
We correctly understood these systems were "brittle" in ways that human reasoning was not.
Finally we have AI systems that are not brittle.... God help us.
... a "small" exercise remaining is to graft such an LLM substrate under the well understood brittle AI reasoning systems we have built over the last 70 years.
I wonder if it is worth thinking of use of LLMs as cheating at the language games we play.