I'd be looking for the person who told wanted the hacking done. I doubt an AI agent does these things without someone instructing them. That would be like seeing a self-driving car go joyriding.
More than likely the instruction was "Fetch this information from the website", and when the agent didn't find that information, it decided that the best way to get it was to hack the site.
If so, it illustrates quite well the lack of common sense in LLMs. A person, especially one with sufficient skill to actually hack a website, would presumably think twice about doing it (considering that it is illegal) for a simple information gathering request.
I wonder if there is any other ways to solve this long-term than to introduce strict liability for model providers...
Edit: An obvious other choice would be strict liability for the operator, but considering how much weird shit LLMs get up to without being asked to, that would get out of hand quickly.
But you see, they never asked a humanoid robot to rob the jewelry store. They just drove robot right next to the store doors, dumped a tools box with hydraulic cutters and diamond saw next to it and instructed an autonomous robot to collect a million dollars until the morning. But they didn't instruct the robot to "rob" the shop specifically, nonono, your Honor. They are honest blokes and never intended to breach the law, it was the robot's intention, see. :)
I think the intent was to get some healthcare statistics, which seems like the kind of thing they would use to test agents; and the agent thought "what is a good way to get the statistics? Hack into Austrlian government and fetch it".
Some people on Hacker News would argue that's what agents should do! I remember the other thread about hacking chess engines, multiple people argued "yeah I want the agent to do that"!
An AI agent can't make decisions, so there's that. But, as for purpose, it's not the purpose of an AI agent we need to fear, but the person wielding it.
Weird, when I tell it "pick between the options you presented the one you think it's best" it does make a "decision". But now you are going to say "nothing was decided here, no more than how a flipping coin decides on what side to fall".
The Linux desktop does need work, but it's not 100% up to speed yet, either. The ultimate UX can be found when the user has 100% control over the experience and we aren't there yet. Wayland won't allow window size/position to be remembered between sessions and too many UX designers think the desktop is just a big phone. Yes, we need work on UX and the first step should be to get Wayland to allow the user the same flexibility we enjoyed with X11. Using Linux with multiple screens means reverting to X11 just to keep from going nuts.
> Wayland won't allow window size/position to be remembered between sessions
It's not on the program itself to restore its position, but on the compositor, who has every right and control to do just that. AFAIK it does work in plasma Wayland.
That's the problem. The compositor can't reliably distinguish windows, and it definitely doesn't know when the data should be deleted in case the same window never opens again.
That sounds great, but I don't fully agree... IMO a compositor should be able to provide a data structure (unchangeable, mostly or fully opaque) that the app can save/restore across sessions.
But then what's the point, what do you do with it that the compositor can't?
Then instead have stable window IDs so the compositor can do its job properly - there is absolutely no need to self-position and outside of certain privileged apps I absolutely don't want random programs to move their windows around.
I feel very lucky in that I bought a handful of SSDs mere months before the prices went insane. RAM, not so much, especially for my new laptop which demands DDR5. It may be a while before I can expand that past the 16GB that came with it.
I dabbled with Flet for a couple of weeks and all I can say is: I wish it were further along in development. Either that, or had more documentation and examples available. I love that it targets all platforms and that it's so easily distributable, but where I ran into problems was with layouts in general and aligning controls with one another specifically. I was ready to drop all other GUI toolkits before I ran into that snag. Perhaps in another year or two I'll take another look.
The layout documentation is a real weak point in Flet. I believe it's essentially as powerful as Flutter, which is also really hard to figure out but at least there's a million blog posts about it and AI understands it. Flet repackages that power (mainly in its Container widget) in a way that's probably more sensible, but if they'd religiously stuck to Flutter's API (or just had really good documentation) then at least you'd be able to figure it out.
This was just business as usual 50 to 100 years ago. People started at the bottom and worked their way up. There were no I've-got-a-degree shortcuts because degrees weren't about training for a job, they were about finding out about the world around us. It's interesting that things are starting to circle around, but I doubt any corporation will drop the degree requirements... unless AI CEO's start rejecting all human applications.
As a high school dropout, with a GED, I can relate. Pretty much all of my education has been OJT and seminars/classes.
It's earned me lots of sneers, over the years, but the people that mattered, didn't care. I spent most of my career at one of the world's best optical equipment companies, working as a peer with some of the top people in the world. It was quite humbling, and extremely gratifying.
I'm trying to find my focus again. After using AI agents for about a year, I've lost my ability to concentrate on anything for more than a few minutes.
I was interested in reading the article but, like so many I find online these days, the white text on black makes my eyes throb to the point of giving me a headache if I don't stop after a few seconds. so I have to give it a miss. I've read that white text on a black background is easier on the eyes, but it's obviously not universal.
reply