3 ms·
How much experience do you have with LLMs exactly? It would be consistent with my experience if Claude stuck in a line of python that just emits a JSON literal
by jeffbee 8d ago
How much experience do you have with LLMs exactly? It would be consistent with my experience if Claude stuck in a line of python that just emits a JSON literal with no justification, potentially buried in a large program where an untrained person might not notice it. I don't even trust them if the output consists of structured data paired with source images from the PDF, because I've experienced LLMs fabricating the source rectangles to match the output. I only use tools like this by asking for programs, because as you note LLMs are good at that, and the verification process consists of tool calls to legitimate PDF manipulation tools so I have some confidence everything is above board. Even then I only do this for hobbies, not anything that matters.
- terminalcommand 8d agoLawyer here. I used to trust Claude as hallucinations are near non-existent now. However for large volume tasks such as due diligence exercises, they still happen. We also tried Legora's tabular review, there were also numerous halucinated provisions in our due diligence exercise.
- rayiner 8d agoJunior associates hallucinate too...
- NateEag 8d agoAnd when they do, you can train them or fire them, and they learn not to do it. LLMs change not a whit, and there's no one to take responsibility for the failure (and thus no way to fix it). As the new variation on the old theme has it, "A computer can never be held accountable, and so very many people are trying to get them make management decisions."
- IanCal 8d agoYou can’t train people to never make a mistake, particularly when doing highly repetitive work like this. You must build your systems to account for that regardless.
- enraged_camel 8d agoYes, exactly. Humans are non-deterministic as well, just in different ways. A tired human can make all sorts of errors for example, regardless of how much training they've had.
- NateEag 8d agoFor sure. But they do learn and improve. The models don't (yet).
- hollerith 8d agoThe models improve in the sense that GPT 5.6 succeeds at things GPT 5.5 fails at. It might be that the models have been improving in this sense faster than a human child improves.
- juiceland 8d ago> LLMs change not a whit, and there's no one to take responsibility for the failure (and thus no way to fix it). LLM output is nondeterministic and humans take responsibility for the failure the same way they take responsibility of a photocopy is too dark.
- NateEag 8d ago> humans take responsibility for the failure the same way they take responsibility of a photocopy is too dark. You mean, they notice it's too dark right after making it, change the settings, do it again, and give you the good copy? Because yes, that's my experience of humans.
- podocarp 8d agoYou can scold juniors and they will learn. You can't scold Claude.
- stevesimmons 8d agoSurely the rate of improvement in new LLM models is the equivalent mechanism?
- eru 8d agoYou can scold Claude. Just doesn't make a difference.
- chrisjj 8d ago... until you hit Claude's risable "model welfare" protection.
- egorfine 8d agoIs it possible to catch those hallucinations using another LLM with a strong fact checker prompt with sources provided in output for human validation?