[eschaton]

How do you square this with the fact that LLM-based tools aren’t actually doing any analysis whatsoever, but are just pachinko machines that produce statistically likely output tokens when given input tokens?