the rm -rf's will continue until morale improves
-
the rm -rf's will continue until morale improves
@neurovagrant@masto.deoan.org I heard they also set it recently so that by default it will be able to run destructive commands without requesting permission
Don’t know why anyone would trust this stuff -
the rm -rf's will continue until morale improves
@neurovagrant I realize that formal verification is a real pain; but I can't help but get the strong impression that software's experiments with downright belligerent levels of dumb empiricism might be going badly.
It's like watching civil engineers overcome with excitement at the velocities achievable under "hold my beer" standards and practices.
-
If you ask a human to explain an action, they think back to what they were thinking at the time. LLMs don't do that, because LLM's don't think.
GenAI engines are chatbots. They generate text. They match your prompt against some subset of the billions of rules they have and the combination of rules that trigger most strongly determines the response, one word at a time.
The only honest explanation for “why this behaviour and not that behaviour” would be “these rules triggered more strongly than those rules”.
-
@neurovagrant In 2005, we had a rookie security consultant doing a penetration test for our client. To demonstrate that the client's app was vulnerable to SQL injection, he ran
drop table users;via injection.Twenty years of progress means we can make the rookie mistakes faster and with fewer employees.
@paco and without ever learning any valuable lessons
@neurovagrant -
@jrdepriest @jztusk @paco @neurovagrant "oh, it's only data exfil, that's not so bad" Some people have very limited imaginations, and need an unequivocal demonstration.
@womble @jrdepriest @jztusk @paco @neurovagrant "Limited?"
-
@jztusk @womble @paco @neurovagrant
Like leaving an empty file called
pwnedon/.@jrdepriest @jztusk @womble @paco @neurovagrant Also fun.
-
@jztusk @womble @paco @neurovagrant
Mission accomplished with the table drop, then?
@jrdepriest @jztusk @womble @paco @neurovagrant Yep. Sometimes that's the only way to make them understand. Or using a loop and ssh to power down an entire rack.
"Are you sure? This is fatal," during a meeting with a client is not an idle threat, and a lot of clients think we're bluffing.
-
@paco @neurovagrant wouldn't that be a criminal offense at that point as well
@rail @paco @neurovagrant Depends on how aggro the client wants to be.
-
@neurovagrant people still haven't learned after all that ransomware. Back it up and/or employ snapshots. I'm a big fan of ZFS snapshots.
@adamhotep @neurovagrant Because they were lucky enough to have usable backups.
Stress "lucky."
-
@Viss @neurovagrant Letsss goooooo
-
@drwho @quinn @neurovagrant so i just went and squinted at the screenshot inside the screenshot
dude doesnt know what a sandbox is.
if the shit can reach your home directory, its not in a sandbox.
-
the rm -rf's will continue until morale improves
-
@neurovagrant also, people still not doing backups/restore testing, no zfs, no jails / containers / other restrictions. Kids these days.
@dch @neurovagrant There are folks who don't even know what removable storage is, let alone file systems or files.
"Just rebuild."
Great. That's the OS. What about the data?
There's probably a "chased by an angry goose" meme to make here.
-
@drwho @quinn @neurovagrant so i just went and squinted at the screenshot inside the screenshot
dude doesnt know what a sandbox is.
if the shit can reach your home directory, its not in a sandbox.
@Viss @quinn @neurovagrant Given his previous business history, this surprises not at all.
-
@Viss @quinn @neurovagrant Given his previous business history, this surprises not at all.
@drwho @Viss @neurovagrant he probably makes more in a month than i will ever make
-
@drwho @Viss @neurovagrant he probably makes more in a month than i will ever make
@quinn @drwho @neurovagrant oh, hes a cryptobro. no wonder
-
@drwho @Viss @neurovagrant he probably makes more in a month than i will ever make
@quinn @Viss @neurovagrant Same.
-
@neurovagrant In 2005, we had a rookie security consultant doing a penetration test for our client. To demonstrate that the client's app was vulnerable to SQL injection, he ran
drop table users;via injection.Twenty years of progress means we can make the rookie mistakes faster and with fewer employees.
@paco @neurovagrant imho, if LLMs are what make developers finally care about cybersecurity, that's the first thing about them one might consider 'good'
-
@rail @paco @neurovagrant Depends on how aggro the client wants to be.
Also depends on how well written your contact with the client is.
-
the rm -rf's will continue until morale improves
@neurovagrant I'm pushed to use claude at work... Three possible takeaways from this:
- Don't use auto mode and review the commands.
- Sandbox claude so that the fallout of such an oppsie is more limited.
- Do nothing and have a claude-free afternoon while I restore my profile.
