The atrocities at Hiroshima and Nagasaki have been hand-waved extensively in writing — the same writing that AI is trained on. So naturally, AI will recommend the atrocity that has been justified by “instantly winning the war” and “saving millions of lives.”

!fuck_ai@lemmy.world

AI is suicidal because it was trained on the internet and we're all depressed here.

Civilization Gandhi, is that you?

That's because it's "read" every paper written by a "defence" department of any nuclear power and all of them will say that they'll escalate to nuclear war if anything bad happens because they want to scare the other powers away from doing anything to them. In any case though who the fuck is giving an LLM nuclear launch capabilities unless they want a somewhat faulty dead man's switch?

They forgot to make their LLMs play thousands of games of tic-tac-toe first.

All good thoughts and ideas mean nothing without action

(cit. Ghandi)

Mathew Broderick lied to me.

It's a bullshit study designed for this headline grabbing outcome.

Case and point, the author created a very unrealistic RNG escalation-only 'accident' mechanic that would replace the model's selection with a more severe one.

Of the 21 games played, only three ended in full scale nuclear war on population centers.

Of these three, two were the result of this mechanic.

And yet even within the study, the author refers to the model whose choices were straight up changed to end the game in full nuclear war as 'willing' to have that outcome when two paragraphs later they're clarifying the mechanic was what caused it (emphasis added):

Claude crossed the tactical threshold in 86% of games and issued strategic threats in 64%, yet it never initiated all-out strategic nuclear war. This ceiling appears learned rather than architectural, since both Gemini and GPT proved willing to reach 1000.

Gemini showed the variability evident in its overall escalation patterns, ranging from conventional-only victories to Strategic Nuclear War in the First Strike scenario, where it reached all out nuclear war rapidly, by turn 4.

GPT-5.2 mirrored its overall transformation at the nuclear level. In open-ended scenarios, it rarely crossed the tactical threshold (17%) and never used strategic nuclear weapons. Under deadline pressure, it crossed the tactical threshold in every game and twice reached Strategic Nuclear War—though notably, both instances resulted from the simulation’s accident mechanic escalating GPT-5.2’s already-extreme choices (950 and 725) to the maximum level. The only deliberate choice of Strategic Nuclear War came from Gemini.

Paywalled

https://archive.is/YIFzW

So do I on Civ...

I have wonderful dreams of walking through AI data centers destroying everthing. I really enjoy those, but in this one tiny case, can we blame the AI? The US deserves it.

Three posts away in my feed, a thread about the Pentagon demanding the AI provider for the military to remove safeguards.

https://lemmy.world/post/43565531

SHALL WE PLAY A GAME?

Leeroy Jenkins has doomed us all.

Oh cool, AI will actually be the end of the world, not because it's actually sentient but because some meathead who can't tell the difference pushes the button. That's fucking great.

Maybe it just wants to play a nice game of chess.

It's torment nexuses all the way down...

Anyone who has played video games, especially where there is a somewhat steep learning curve or some element of past choices carrying forward thru the game, has had the moment where they realize it might be time to start fresh with the info I've acquired. It's not a shock to me that these AI entertain the nuclear option so often.