Also, courts record charges which are dismissed due to having no evidential basis whatsoever and statements which are deemed to be unreliable or even withdrawn. AI systems, particularly language models aggregating vast corpuses of data, are not always good at making these distinctions.

That is a critical point that AI companies want to remove. _they_ want to be the system of record. Except they _can't_. Which makes me think of LLMs are just really bad cache layers on the world.