Hacker News new | ask | show | jobs
by bombcar 17 days ago
90% of the time I ask a question of a coworker that could be googled or clauded what I’m actually asking for is their confirmation that they agree with the answer. So use the AI, but at least read the reply and/or reword it so it’s clear that you agree.
2 comments
onemoresoop 17 days ago
Maybe they don’t wanna take responsibility for that answer?
link
mrweasel 17 days ago
In our case we're completely allowed to use AI, but if you do, you carry the responsibility for it's output (not in a legal sense obviously). So if your LLM says "Sure, go ahead and run that code" and it deletes the production database, that's your fault, just as much as if you reviewed it manually and said the same thing.
link
bigfishrunning 17 days ago
> you carry the responsibility for it's output (not in a legal sense obviously)

why not in a legal sense? If someone asks me what cleaning supplies are safe to mix, and i just ask some chatbot, don't vet the output, copy the response, and they end up poisoning themselves, am I not responsible?

If I'm a lawyer, and pass unvetted AI legal advice to my client, and they go to jail, should i keep my license?

link
mrweasel 16 days ago
Typically if you develop a software product for a company, the company resumes responsibility. No one would develop anything for any high risk business, if they were to assume personal responsibility of the end result. LLMs don't really chance that, and any serious bug would be an error of process.

Developers are perfectly capable of creating dangerous or expensive mistakes even without LLMs. If we accept that LLMs are just tools, the developer should be no more responsible that if they choose to use Visual Studio over Vim and Visual Studio refactors something wrong.

Theres's then the question of gross incompetence, and were the developer could be fired or perhaps even sued by their employer, the same as if they hadn't used an LLM.

However the case I had in mind was in regards to the legality of the way the AI models acquire and reproduce code. We're not held personally responsible for any license violation created by the tool.

link
onemoresoop 16 days ago
The personal responsibility is that you can lose the job which may not be as bad as getting sued but right now the market is quite tough so…
link
dapperdrake 17 days ago
So, just a fake "yes"?
link