> The first year that commercial LLMs became available, a lawyer could plausibly claim they didn't know any better that an LLM would fuck things up

Is that the case? As an officer of the court, lawyers have a duty to be truthful. Putting forth an argument you didn't even read would seem to be abdicating that duty regardless of the novelty of the tool.

In 2023 we didn't know that LLMs would make things up even if you asked them to limit themselves to actual facts/laws/etc.

Does that change the duty of the attorney? Paralegals can make stuff up too. Chatgpt was released in Nov 2022 and the first case of sanctions was brought in early 2023 so it doesn't seem like it was ever acceptable.

A paralegal is a human, and specifically a human that has been trained to research and not make things up.

You're right, I had my years off by one...common programming error...In 2022, people thought that LLMs were tools like Word, or a calculator. The realization that they were black boxes that could make things up or return different results for the same prompt wasn't well known back them.

My point was the first case for sanctions against a lawyer using AI was only two months after the debut of ChatGPT which undermines your thesis. You just keep repeating yourself as if that makes it so.

I’m talking about mechanically editing a Word document, not generating complaints. You can tell it to CRUD paragraphs, images and tables verbatim.

this is a common misconception, but in my experience including these kinds of instructions will often work, but not every time. this means that every document still needs to be carefully checked, and it's usually easier to just do the work oneself