The virtue of contanarized environments.
Successfully uninstalled! test passed.
Why delete the files when I can just delete the computer? 🔥🔥🔥
“This AI is not very intelligent”
Keep going dude, you are almost there.
but super intelligence
Super dooper intelligence.
“i got ai to generate code i couldnt read, and ran it anyway”
moronic clown behaviour
Right? Moral equivalent of “I played with fire and burned my house down”
But didn’t you see? They thought it could never happen to them! How could they have known they weren’t magically immune to this very common issue where everyone keeps saying “I thought it could never happen to me but it did”?
A lot of these idiots are giving the AI permission to directly run code on their computer.
Never mind ‘couldn’t read’ – they’ve never seen the code at all.
It doesn’t help that the latest version of Google antigravity is not a vs code proprietary fork like in the beginning but a giant empty window with just a prompt, that hides everything not to scare the prompters
What changed? Can I see and manually edit the full code and not the diff? Nope
(Disclaimer: I didn’t actually use the program, my assumptions are based on screenshots, I don’t know if there’s a setting to enable full editor view)
Imagine that the first race, facing the Reaper threat and having no idea how to defeat them, sits down and designs a trigger guard. And that’s it. Then they bury the plans for the trigger guard and they die. 50,000 years later, the next race is getting pulverized. Before they die, they find the plans for the trigger guard. They have no idea what it’s for or what it does, but they design a handle to go with it. They add it to the plans, and re-bury them.
And so it goes. 50,000 years. A safety mechanism. A rifled barrel. A magazine. A rear sight. The trigger. A front sight. A muzzle. An ejection port. Nobody knows what any of this does.
Then Shepard & Co come along. They follow the plans, which builds a Glock 17 pistol. Admiral Hackett points to the chamber. Something goes in there, but we don’t know what it is or what it does.
Then you meet the Star Child, who just happens to be a 9mm bullet, which miraculously is a perfect fit for this pistol, even though the people who built it have no idea what a bullet is or what it does.
Then the Star Child explains that the next step is to put the bullet in the chamber, aim the weapon at your foot, and pull the trigger. That’s how you “win”.
First time at the end I immediately attacked the star child.
Turns out that’s also an end.
At this point it’s pretty easy to sandbox your agent so it can’t do this. Even a trivial docker container with a shell script would have prevented this.
While i understand you would think a vendor like Google would do sandboxing by default, the industry standard is that your agent gets your user permissions in full. Your external drive doesn’t need escalated permissions to wipe, and the AI companies don’t care at all about your security.
The T in AI stands for trust.
Sandboxing wouldn’t have helped as he asked to create an executable that deletes files. It’s not the agent that deleted the files but the “uninstall” routine in the compiled program that deleted everything.
With such a generic prompt I wonder what kind of mess it has generated.
Probably such a program didn’t need an “install” or “uninstall” routine, as simple programs are portable, don’t need to leave libraries on the disk. He even didn’t know that uninstall and install are the last thing you do in a program, and It’s not integrated in the code, but it’s packaged after the build with third party stuff like nsis or innosetup
Clearly, they forgot to add “make no mistakes” to their prompt.
This AI is not very intelligent.
Writers’ rooms everywhere are in shambles as reality rapidly approaches peak comedy.
Implying writers rooms aren’t a complete fucking shambles
Knows about similar cases happening to others
Assumes they’ll be fine and freely uses slopbot anyhowLooks like the bot’s not the only thing lacking intelligence here.
I’m always suspicious of screenshots of posts that cut out the date and author. If anyone is curious, this is real, source is here, archived here if it gets deleted.
For bonus points, while searching I also came across this one from last year. Different scenario but similar outcome, where it wiped a guy’s drive for a routine project clean up
That last sentence… So much lack of self awareness… People trusting black boxes yet thinking they are geniuses. Hey moron, antigravity is only very dangerous because you’re very stupid.
Also “This AI is not very intelligent”
Meanwhile, everyone and their dog is vibe coding applications for various homebrewed consoles. And the creators invariably get all huffy when you aren’t enthusiastic about installing a program they don’t understand and has unknown amounts of testing.
“I told the AI to test it and it reported that everything worked exactly how I wanted”.
Oh boy! A golden opportunity to trot out this classic again.
(Myth 2’s initial version, which was released to retail in a time when patching games over the internet was not quite as viable as it is today, let alone distributing them via the internet, had a bug which would delete the entire contents of your machine’s C: drive if you uninstalled it when installed to any directory other than its default. Truly, to err is human. But to really fuck up you need a computer.)
Ah, the times where a massive bug that wipes the users hard drives can’t be patched and you need to almost bankrupt the company to recall all the printed copies
These slopstitutes really think they’re developers but they haven’t heard of the 3-2-1 backup method
Three copies, two different media, and one should be stored in sight at all times, right?
I always consider the “1” to be stored off site
That is how the saying goes. Never heard someone said in-sight. That is super silly. It’s off-site in case of massive disasters like the place burning down or nearly burning down but getting damaged by water used to douse the flames.
