Comment by not2b

You are arguing with your own straw man interpretation of the article. It isn't talking about all possible uses of LLMs, but focusing on specific uses now being proposed, to use ChatGPT and its possible successors instead of search.

You ignore his points about how achieving really good compression requires learning structure in the data that starts to amount to understanding: if you can understand the rules of arithmetic then results of arithmetic operations don't need to be saved, they can be reconstructed accurately. You mistake his pointing out the arithmetic errors ChatGPT makes for a claim that they are fundamental to LLMs, even though in the same article he points out that understanding arithmetic aids compression. And then you say that his article shouldn't have been published.

I will not say that your comment shouldn't have been published, because that is impolite and against the rules.