something they know isn’t correct ALL of the time
Neither are humans…
Humans can be held accountable though. Which means they would rather say “I believe this is because …” than “you are absolutely right! If you remove the voltage regulator from the power supply, it will provide more power!”. And if they don’t, they can be fired or put in prison, where they can stop doing harm.
If your LLM repeatedly lies to you, there’s nothing you can do about it. Maybe try another LLM, but there’s not many of them and they all lie.
I think the perspective of moral responsibility (“lie”, “held accountable”) is not a useful one here. Punishment is one way of discouraging humans from making mistakes but it can’t prevent mistakes entirely. My field is software development and in software development there is the expectation that everyone, even the best human developer, makes mistakes relatively often, and there are frameworks for managing those mistakes. AI can fit into those frameworks.
WhatsHerBucket@lemmy.world 2 days ago
I expect Joe at the cash register might occasionally count his drawer wrong. However, he’s not going to randomly give me made up shit.
ArbitraryValue@sh.itjust.works 2 days ago
I agree that AI is inhumanly good at presenting wrong information confidently and convincingly.