[renewiltord]

A personal target I have is that I compare my intelligence to an LLM by seeing if it correctly interprets things people say that I have trouble with. In this way, I can see where the LLMs are beating me. I had no trouble with this headline but had I had trouble, I would conclude that ChatGPT-3.5 is smarter than I am on headline interpretation, because it interprets the headline concordant with the story first time.

In this model, I would judge myself as less able to extract information from other humans than ChatGPT. I believe that the ability to extract signal is a good marker of intelligence. Low-spec intelligences often require high-precision: they are more like formal language parsers than natural language parsers and are therefore less sophisticated.