Post

Most people mistakenly assume "because computers are good at math and word problems, ChatGPT should be". Ironically, LLMs are *worst* at those: they can't do iterative tasks (like math), and they don't "see" words or letters (only tokens). They're best at logic and tool usage.

It's kind of the whole point of AI: the goal is *artificial intelligence*, aka to gain capabilities like humans. So talking about things that computers are good at doesn't come into play - except where LLMs are allowed to, like humans, use tools to solve tasks (which they excel at).

The problem with that angle is it’s predicated on limited ideas about a job’s value. My “dumbest coworker” was very friendly & made people feel at ease. She could not however use Excel or problem solve at all. Still she had important skills I didn’t.