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.
And they don’t have cognition at all. They do not, and can not, think like we do. Maybe some day we will learn to make real AI, these LLM’s are not it. It’s a cheap trick intelligence,.
I think the emp is pretty limited to the blast zone in frying electronics. The fallout from a weapon spreads around the world, circling in the winds countless times dropping dust everywhere, but the emp is localized to more around the area of physical destruction but not sure exactly.
The Neutron bombs, not entirely sure in physics how that works, but they produce no actual blast that causes physical destruction so much and just kills everything.
Oh, how far from the blast and how does it mess them up do you know? I should know that I guess I just heard about the emp, and not sure how a neutron bomb would affect electronics either.
No, that I can’t answer — it would depend entirely on the level of fallout and where it happens to land.
You would need to be able to perfectly, and I mean perfectly, predict weather months in advance in order to prepare accordingly.
The reaility is that for an AI, or rather an AGI, to make the choice to launch nukes would require them to reach a point where they accept the potential loss of their own ‘life’ in exchange for whatever value a nuclear war might hold. I struggle to believe that a ‘true’ AGI would make that choice. There are far too many variables to control in comparison to a biological agent, one that likely would not affect a machine.
Now, a modern AI making that choice? Absolutely possible, the things are fucking crazy with literally no concept of what life is.
An AI can easily start nuclear war, as can a human.
The only thing preventing a nuclear disaster are all the institutional measures limiting its accessiblity.
If you gave a single human (or a single AI) access to a magic no-strings-attached ‘Send a Nuke’ button, either the human/AI is the second coming of Jesus Christ, or a nuke will befall some unlucky portion of the population sooner or later. Bonus points if people can talk to the AI or if access to the button is hereditary.
If you think computers aren’t affected by radiation or nuclear winter I’ve got some bad news about where their power comes from and what the main principle of electricity is
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):
No human has ever deployed tactical nukes against a nuclear capable enemy.
“no human” but Machines would, since they are unaffected by nuclear winter and radiation.
And they don’t have cognition at all. They do not, and can not, think like we do. Maybe some day we will learn to make real AI, these LLM’s are not it. It’s a cheap trick intelligence,.
Radiation absolutely fucks electronic components
The electromagnetic pulse caused by a nuke would pop resisters too. AI would more likely use biological means to get rid of us.
Like heating the planet another degree and starving us out of existence by killing off biodiversity until the crops die out… Like they’re doing now?
(I say “Us” when I just really mean the 99% of people that haven’t got self sufficient underground complexes)
Assuming AI would care about itself and not just “solving the problem”.
Yeah, these doom scenarios require cascading assumptions and no real answer, except maybe “don’t”.
I think the emp is pretty limited to the blast zone in frying electronics. The fallout from a weapon spreads around the world, circling in the winds countless times dropping dust everywhere, but the emp is localized to more around the area of physical destruction but not sure exactly.
The Neutron bombs, not entirely sure in physics how that works, but they produce no actual blast that causes physical destruction so much and just kills everything.
I repeat, radiation absolutely fucks electronic components. I am not talking about an emp, I am talking about radiation.
Oh, how far from the blast and how does it mess them up do you know? I should know that I guess I just heard about the emp, and not sure how a neutron bomb would affect electronics either.
No, that I can’t answer — it would depend entirely on the level of fallout and where it happens to land.
You would need to be able to perfectly, and I mean perfectly, predict weather months in advance in order to prepare accordingly.
The reaility is that for an AI, or rather an AGI, to make the choice to launch nukes would require them to reach a point where they accept the potential loss of their own ‘life’ in exchange for whatever value a nuclear war might hold. I struggle to believe that a ‘true’ AGI would make that choice. There are far too many variables to control in comparison to a biological agent, one that likely would not affect a machine.
Now, a modern AI making that choice? Absolutely possible, the things are fucking crazy with literally no concept of what life is.
An AI can easily start nuclear war, as can a human.
The only thing preventing a nuclear disaster are all the institutional measures limiting its accessiblity.
If you gave a single human (or a single AI) access to a magic no-strings-attached ‘Send a Nuke’ button, either the human/AI is the second coming of Jesus Christ, or a nuke will befall some unlucky portion of the population sooner or later. Bonus points if people can talk to the AI or if access to the button is hereditary.
If you think computers aren’t affected by radiation or nuclear winter I’ve got some bad news about where their power comes from and what the main principle of electricity is
What you’re thinking of is Terminator