The example of ChatGPT performing math is one of the pieces of evidence that convinced me of this post. ChatGPT, you many know, is inconsistent regarding math. Sometimes it gets things very wrong. This, however, is actually an argument that ChatGPT is performing math closer to how humans perform math. Humans often make simple errors, especially as children, whereas calculators do not make errors. GPT is using linguistic templates for solving math problems, and is doing so inconsistently. That is also how humans do math.