3 ms·
Were you using 3.5 or 4? I've used 4 and been very impressed with it. For business it's been great, because the language is usually clear and unambiguous. Fo
by JamesBarney 4y ago
Were you using 3.5 or 4?
I've used 4 and been very impressed with it.
For business it's been great, because the language is usually clear and unambiguous. For employment contracts it's been not nearly as helpful because the non-compete language is meant to be vague and ambiguous, and relies on a lot of case law around what is and isn't enforceable.
- gamblor956 4y agoBoth. 3.5 was a non-starter. 4 is improved, in the sense that it produced stylistically correct looking fluff but still offered the kind of substantive legal nonsense that would get a human lawyer disbarred for gross incompetence: it still makes up laws, and cases, and facts. This is because ChatGPT doesn't have knowledge or understanding of the law, or facts, or the consequences of its output. It just matches linguistic tokens probabilistically. Those who rely on ChatGPT for legal advice will make some human lawyers very rich some day when they have to hire real lawyers to clean up ChatGPT's mistakes. This isn't to say that AI is incapable of providing legal advice; but it's not going to be an LLM (though it might be a KbLLM).
- JamesBarney 4y agoWhat were the really simple legal questions it got wrong?
- gamblor956 4y agoIt couldn't handle basic wills, or pre-nups, which are literally just templates in the most basic case. It couldn't explain how to get out of a traffic ticket. Oh, and it wasn't clear on how Miranda works. If it can't handle easy stuff which is literally just boilerplate, it can't be trusted to handle more important stuff where actual judgment or creativity is required.