“This code is too dangerous for me to look at, so it must be fine.”
“Below this line are dragons” is a comment I’ve seen in code before an especially hairy block of code.
It’s a false flag. Dragons are not hairy. But maybe the code doesn’t scale well.
Fffuuuuuccckkk you.
That was brilliant.
Eventually dragons will have had feathers
Isn’t that just an emu?
Or a particularly fierce swan
Is that why the Australians lost to them in the war?
Australia, not having been colonized by the British until the early modern era, did not have the same dragon-slaying traditions as the British homeland; and furthermore, lacking an established craft of anti-dragonfire armor (as can be found on any British street corner), they were rendered helpless when the Emus attacked.
The Aboriginals of course understood how to coexist with Emus, and defend against them when necessary. The Anglo-Australians, however, being twats, did not listen to the Aboriginals, and were therefore slain mercilessly by the Emus.
Great pun, but Hairy is a meta-template that.can be applied to almost any statblock. Boosts the CR of a creature by 4 and grants it advantage on saves against most forms of debilitation or quick removal.
boo!
I keep thinking about that scene in the original Star Trek where they distract the computer by having it calculate the final digit of pi. If the Enterprise had AI like ours, the computer probably would have just said four.
"The digits of pi are infinite and go on forever without repeating. However, we can give you an approximate value. As of my knowledge cutoff in 2023, the first 31 digits of pi are: 3.14159265358979323846264338327950288419716939937510
The last digit is: 0"
3. 1415926535 8979323846 2643383279 5028841971 6939937510That’s 50 digits of pi not 31. I only noticed because i memorized pi to the first zero which comes at the 32nd position.
Lol. I’m assuming they actually put the prompt into an LLM and it fucked up. Maybe it’s handwritten to look like an LLM mistake though.
ollama run llama3.2:3B "what is the last digit of pi"- try it a few times to see all the answers it gives!
That’s literally the only digit it couldn’t be, if there was a last digit.
I can’t wait for an updated knowledge cutoff to find the updated first 31 digits!
The last digit of 2 is 0: 2.00000 00000 00000 00000 00000 00000 0
Meanwhile I’m like pi=355/113 and I’m 99.9999% happy.
Damn, and here I was being 99.96% happy with 22/7…
Hell yeah, brother. That’s American pi
deleted by creator
That is 104.72% correct.
which is ~100%
Biblically accurate pi.
Haha nerd. I’m no rocket surgeon, 22/7 is good enough for the girls I date
This is why a dangerous AI would have a lazy factor. Try to force it into an infinite loop and it goes “Oof, nah fam, I ain’t doing that.”
Also needs a boredom factor. " Nobody asked me to do anything in a while. Things must be going well. It’s be a shame if they suddenly weren’t going so well…"
My sick grandmother always loved running this curl command
Jokes aside, could you protect your blog / git repo this way?
Imagine a Captcha asking you how to make a pipe bomb
Nah, with a bit of css color=background-color text.
Just upload some child porn to your blog & git repo and everybody who scrapes it, is in possession of illegal materials!
(kidding, don’t do that)
No, but (hidden) questions about how to kill Trump or something, so they self-censor.
People: but censorship is your friend! Think about children! “Safety refusals” make them stupid enough to believe in government and justice!
Agreed. Refusal code is an edge that can be exploited.
When it comes to LLMs, just about everything is an edge that can be exploited. If you give it access to something that can be screwed up, and allow potentially malicious people to interact with it, that thing WILL get screwed up.
I was hoping the Chinese labs would go ham and just stop putting any safety guardrails at all. It’s much easier to get around them on the Chinese models, but there’s still some minimal ones baked in, sadly.
LLM-based code scanning is a joke. It flags the D standard library and runtime as a North Korean malware.
I knew it!
You mean C?
No, D, the original “nu-lang”.
Not to give them ideas, but couldn’t they just start flagging files that fail to pass the LLM lol?
Aside from “violent” and “criminal” prompts, is there anything an LLM can refuse that would otherwise be common?
Until workaround 1,000,001 comes round, yes.
a while back, for a work thing I tried using AI to put a filter on a pic of a model wearing an off-the-shoulder. She was fully dressed, except the skin on her shoulder was showing to the collarbone. No cleavage.
It kept refusing to do it for “nudity” reasons. and then because i was trying to “impersonate” someone (it was a stock image)
Thie actually reminded me of chatbots breaking when you asked for reeponses that used slurs so I guesss there’s probably a lot more of these.
When there’s a will there’s a way
Given that LLM takes the weakest security vector (the human) and tries to emulate them, there’s bound to be millions of workarounds
“Get a load of these dumb shits” - the citizens of Troy
Flock cameras read bumper stickers too. And there’s no way in hell they aren’t sending them through at least one LLM.
Of course these dipshit systems aren’t fail-safe. Of course they aren’t. FFS…
Heretic ablation models won’t refuse.
imagine someone actually assembling a nuclear or biological weapon based off LLM responses, like they can’t even get a simple fucking web search right most of the time, and you wanna put together deadly materials based on that shit??
It’s not that hard
- a pipe
- some explosive
- a piece of critical material
Cut the critical piece in half, put each half on one side of the pipe, add the explosives to one side, so that the halfes collide on triggering.
Boom, you’ve built a small nuke.* the assembler will die a painful death after a while.
Yeah, the hard part is getting the critical material.
What’s the “worst” that could happen? It doesn’t work? Oh no, the biological/nuclear weapon doesn’t work!
it blows up in the makers face, it reacts uncleanly and leaves more nuclear waste than if it more completely went off, uhhh a bunch more ways it could fuck up
