It's mostly useless for anything with numbers or other hard facts. Sure, 80% of the numbers would be correct, but we can never know which 80% and would need to

It's mostly useless for anything with numbers or other hard facts. Sure, 80% of the numbers would be correct, but we can never know which 80% and would need to manually verify everything. Not that Google is better in complex tasks). LLMs are mostly useful for producing some vague non-factual business speak, or some abstract imprecise media. This is natural result of the architecture of these programs.