The only safe system is one that is disconnected from other connections physically, then shielded, then with very limited access. Even then, humans are still the biggest risk. I'm sure they're doing the exact opposite.
By definition LLM models need massive external input in order to improve. So they can't really be disconnected. Top that off with them only being useful when you can interact with them from many or remote locations, and there's just no way to really keep them secure. They need massive communication to accomplish anything useful, and there's no real way to keep massive communication secure.
You don't have to improve them out in the field. Just collect metrics on their behavior and train a central model on that data, then you upgrade the local models on each unit when they are brought in for maintenance. I'm simplifying, of course. And terrified.
That's the thing, the local models aren't going to have the processing power to really beat deterministic systems. So you're going to need comms at some point to really get any kind of edge. Otherwise dumb systems get the job done for pennies, while you're having to produce high end chips to handle local processing and back end training, just to significantly out perform a garage door trip light. Yes, an exaggerated comparison, but the point holds.
Did we miss the bit where Trump literally gave MechaHitler's handler a government appointment to rape the PII systems of the US Government for the purposes of "reducing spending" stealing PII information to spy on citizens and immigrants?
Two astronauts are in frame. One if them is facing the earth from afar. The other astronaut is behind the first, pointing a gun at the astronaut, who is asking the question "The US military is team MechaHitler now?"
Musk shouts the loudest about AI must lead to Startrek (UBI/taxes on megaprofits) instead of Skynet. His actions are all in on Skynet and Naziism, because that is what pays the most.
This is going to be a new sabre rattling technique in the future.
If you don't back down we're going to turn our entire military over to AI and it might do anything.
This is either going to turn out to be nothing or an utter catastrophe and there's no option in between. I suspect most generals in the military are bright enough to not give it control over anything too dangerous though.
55 Comments
teft@piefed.social · 56 pts · 200d
How long until these dipshits accidentally publish state secrets on twitter? Taking bets now.
Freefall@lemmy.world · 40 pts · 200d
They don't understand how anything works. They thought USAID was a purely humanitarian effort, ffs.
a_non_monotonic_function@lemmy.world · 2 pts · 199d
Oh, they just didn't give a shit.
nfreak@lemmy.ml · 4 pts · 200d
2 weeks tops
toiletobserver@lemmy.world · 45 pts · 200d
The only safe system is one that is disconnected from other connections physically, then shielded, then with very limited access. Even then, humans are still the biggest risk. I'm sure they're doing the exact opposite.
LuxSpark@lemmy.cafe · 16 pts · 200d
What if the system is connected to the fattest shit pipe in the world X?
toiletobserver@lemmy.world · 8 pts · 200d
Rainbows and unicorns obviously
takeda@lemmy.dbzer0.com · 10 pts · 200d
They are trying to use LLM chat bot to control military equipment. What do you think?
Skyrmir@lemmy.world · 5 pts · 200d
By definition LLM models need massive external input in order to improve. So they can't really be disconnected. Top that off with them only being useful when you can interact with them from many or remote locations, and there's just no way to really keep them secure. They need massive communication to accomplish anything useful, and there's no real way to keep massive communication secure.
raccoon@sh.itjust.works · 1 pts · 198d
You don't have to improve them out in the field. Just collect metrics on their behavior and train a central model on that data, then you upgrade the local models on each unit when they are brought in for maintenance. I'm simplifying, of course. And terrified.
Skyrmir@lemmy.world · 1 pts · 198d
That's the thing, the local models aren't going to have the processing power to really beat deterministic systems. So you're going to need comms at some point to really get any kind of edge. Otherwise dumb systems get the job done for pennies, while you're having to produce high end chips to handle local processing and back end training, just to significantly out perform a garage door trip light. Yes, an exaggerated comparison, but the point holds.
quick_snail@feddit.nl · 31 pts · 200d
This is how we get classified documents leaked to the public. Please proceed..
triple_entendre@infosec.pub · 12 pts · 200d
War Thunder updates incoming
NotMyOldRedditName@lemmy.world · 4 pts · 199d
They're gonna be so stupid about this its 100% going to somehow end up in training data.
AlecSadler@lemmy.dbzer0.com · 23 pts · 200d
kinda hope this leads to grok+military killing Elon
northernlights@lemmy.today · 2 pts · 199d
I could imagine an AI going "my primary directive is protecting this country. Therefore INITIATE DRONE STRIKE ON ELON MUSK'S LOCATION"
gravitas_deficiency@sh.itjust.works · 17 pts · 200d
It’s gonna be so fun to prompt engineer TS/SCI/Confidential data out of this thing. I’m curious to see how long the first exploit is gonna take.
pinball_wizard@lemmy.zip · 4 pts · 200d
I hope there's even time to set up an informal betting pool. I worry the first breach will be too quick to even makes bets on.
gravitas_deficiency@sh.itjust.works · 4 pts · 200d
Tbh if polymarket doesn’t already have a pool for this, I’d be a little surprised
LodeMike@lemmy.today · 2 pts · 200d
Anything redacted will be permanently unredactable if put into training data.
gravitas_deficiency@sh.itjust.works · 3 pts · 200d
That is precisely my point
FreddiesLantern@leminal.space · 16 pts · 200d
Oh Well, back to burning Teslas then I guess.
StupidBrotherInLaw@lemmy.world · 14 pts · 200d
When did we stop?
FauxLiving@lemmy.world · 15 pts · 200d
I can't forsee anything going wrong with giving Mechahitler access to military secrets.
jarvis@lemmy.world · 14 pts · 200d
Wait... the US military is team MechaHitler now?
Zerush@lemmy.ml · 16 pts · 200d
Yes, under Trump commands. Good old ChaosGPT times 💀
deliriousdreams@fedia.io · 4 pts · 200d
Did we miss the bit where Trump literally gave MechaHitler's handler a government appointment to rape the PII systems of the US Government for the purposes of
"reducing spending"stealing PII information to spy on citizens and immigrants?Sunflier@lemmy.world · 3 pts · 200d
Always has been?
FLP22012005@lemmy.world · 3 pts · 200d
Two astronauts are in frame. One if them is facing the earth from afar. The other astronaut is behind the first, pointing a gun at the astronaut, who is asking the question "The US military is team MechaHitler now?"
UniversalBasicJustice@quokk.au · 1 pts · 199d
🌍🧑🚀🔫🧑🚀🌚
humanspiral@lemmy.ca · 2 pts · 200d
Always was, son. Nazis won ww2, just not the German ones.
green_red_black@slrpnk.net · 12 pts · 200d
LaunchesKayaks@lemmy.world · 8 pts · 199d
War Thunder is gonna go absolutely wild with this
echodot@feddit.uk · 2 pts · 199d
War thunder needs to add UFOs, so we all finally get to find out what's going on at area 51
ThePantser@sh.itjust.works · 8 pts · 200d
Well it's been shit y'all. But this sounds like the end is nigh.
green_goglin@thelemmy.club · 7 pts · 200d
The US is beyond cooked.
answersplease77@lemmy.world · 7 pts · 200d
unhinged mode with the nuclear code? what would go wrong
wizardbeard@lemmy.dbzer0.com · 6 pts · 200d
Kinda expected this, after DOGE. Still sucks.
SCmSTR@lemmy.blahaj.zone · 6 pts · 200d
Somebody
Anybody
Ensign_Crab@lemmy.world · 6 pts · 199d
I've seen this movie. The WOPR was smarter and never called itself "mechahitler."
buddascrayon@lemmy.world · 6 pts · 200d
It seems Elmo and Trump kissed and made up and are now wiping each other's asses again.
Zerush@lemmy.ml · 6 pts · 200d
In ships and bases with Nuclear weapons, what can go wrong?
humanspiral@lemmy.ca · 5 pts · 200d
Musk shouts the loudest about AI must lead to Startrek (UBI/taxes on megaprofits) instead of Skynet. His actions are all in on Skynet and Naziism, because that is what pays the most.
FreddiesLantern@leminal.space · 5 pts · 199d
Soldier on the field: “Delta team in position! Send orders and coordinates!”
Pentagon: “We receive you Delta team, standby for transmission”
“Ok Grok, send them the orders and coordinates”
Clanker: “Beep beep. Buy a blue check mark and/or Teslur”
Soldier: “Instructions unclear sir!!!”
Pentagon: “Please clarify how they are unclear soldier?”
Soldier: “It’s AI generated porn of your mom sir?”
Pentagon: “GODDAMNIT ELON!!!!!”
blarth@thelemmy.club · 5 pts · 199d
Holy shirt, I figured it out. THIS is the bad place!
echodot@feddit.uk · 4 pts · 199d
This is going to be a new sabre rattling technique in the future.
If you don't back down we're going to turn our entire military over to AI and it might do anything.
This is either going to turn out to be nothing or an utter catastrophe and there's no option in between. I suspect most generals in the military are bright enough to not give it control over anything too dangerous though.
Raiderkev@lemmy.world · 3 pts · 199d
Hey grok, please fwd all this information to the FSB
tanisnikana@lemmy.world · 3 pts · 200d
No part of this will go badly at all.
SethTaylor@lemmy.world · 3 pts · 199d
Shall we start the bidding war?
Who will pay Musk the most for all of America's secrets?
LoremIpsumGenerator@lemmy.world · 2 pts · 199d
Wait until your military replaces humvee with cybertruck. Jeez what a downgrade
TheTechnician27@lemmy.world · 2 pts · 200d
Relevant Jimmy Davis YTP
SpankyDoodle@eviltoast.org · 2 pts · 199d
Oh god no! We’re dead.
lechekaflan@lemmy.world · 1 pts · 198d
Putin would find that absolutely entertaining.
dogs0n@sh.itjust.works · 1 pts · 199d
(With the CSAM settings maxed out ofc)
echodot@feddit.uk · 2 pts · 199d
As if it has a setting. You can't turn that feature off.
ComradePenguin@lemmy.ml · 1 pts · 197d
Can't they use Grok to handle the controlled release of the Epstein files instead?