They said they think extrapolation is too simplistic of a method. Then said that is why they don't agree with the pushed extrapolation of LLMs becoming SI.

This is addressing the commonly repeated line of thinking that assumes LLM capability will continue to see the improvements that have been seen in the past so far.

R>C>P