[MarleTangible]

I think it's about owning the consequences of one's own actions.

[ktm5j]

If you read the thread the guy does own up to his actions. He actually sounds like a nice guy who admits he made a mistake. He seems more interested in preventing this kind of thing from being possible than he is interested in dodging blame.

I'm happy the guy got his data back.

[MarleTangible]

Sorry, I was responding to the comment, not the article.

> Distinction without a difference.

[traderj0e]

Yeah but it's funnier to blame the AI. And when the "real coders don't use AI" people get pissed off at the joke, you double down instead.

[MarleTangible]

Back in early days of personal computers, the instructor told us that a computer just does what you tell tell it to do, though not always what you thought, and making a mistake thousands of times a second hurts a lot.

[traderj0e]

That's still how the intro programming or CS classes start, "computer is a fast idiot" and some examples of it taking things overly literally

[newsoftheday]

Does that mean the prompt should include: "...and don't delete my production database."?

[nemomarx]

If the agent didn't have delete permissions, or was sandboxed dying other way from your production database, that would handle it. So not running it that way is a decision someone made

[bee_rider]

It means people have to read the commands that they are generating before executing them.

[unregistereddev]

Just in case this isn't hyperbole, no. It means an LLM should not be given that much privilege and that you are responsible for reviewing the tool's output and approving its actions.

[saghm]

"But wait, the user probably just meant that I shouldn't delete the database itself. Removing all of the rows in the table is fine"