Back when #ChatGPT was the hot new thing, I tried asking it a question about a topic I'm fairly familiar with, California criminal law. The question was slightly technical, but it's the sort of thing that every brand-new baby lawyer in every District Attorney's or Public Defender's office in the state would be expected to know. ChatGPT got it magnificently wrong, although it eventually agreed with me about what the answer should have been after I did the Socratic-method-law-prof thing for a while. So today, having read the news about #DeepSeek, I though I'd try the same experiment. Unsurprisingly, the result was the same, but it's somewhat surprising that DeepSeek screwed up in precisely the same way ChatGPT did, including quoting the exact same wrong statute. No idea why that is -- perhaps someone who understands LLMs better than I do (a low bar) could explain -- but same moral to the story: don't trust these tools. Both chats are reproduced here:
