Shall we play a game? My AI nuclear simulation
1–10 of 213 posts
Re: Shall we play a game? My AI nuclear simulation
#2It's good when it becomes clear that a tool is dangerous in a certain way. Like it's good when people show you through their behavior that they can't be trusted
Always use a sawstop if you have a circular saw and never trust an llm with any problem where ethics or trust is relevant.
Re: Shall we play a game? My AI nuclear simulation
#3I love seeing the plot lines of The Terminator playing out in real life.
Re: Shall we play a game? My AI nuclear simulation
#4I love seeing the plot lines of The Terminator playing out in real life.
I was thinking more War Games, but I suppose your example follows logically from mine.
Re: Shall we play a game? My AI nuclear simulation
#5It's good when it becomes clear that a tool is dangerous in a certain way. Like it's good when people show you through their behavior that they can't be trusted Always use a sawstop if you have a circular saw and never trust an llm with any problem where ethics or trust is relevant.
Sawstops are expensive and they don't stop kickback, they are the power tool equivalent of alignment IMO.
Don't forget your riving knife and if you don't learn proper technique, you're gonna have a bad time eventually. This applies to AI as well.
Re: Shall we play a game? My AI nuclear simulation
#6I love seeing the plot lines of The Terminator playing out in real life.
WarGames is what they are more-closely referencing (not that it negates your comment in any way).
I just rewatched it a week or so ago and it really took on a whole new light with the advent of LLMs. When I watched it last I knew that computers couldn't do the things portrayed in the movie. Now? Well not exactly in the way it happened in the movie but a whole lot closer.
I wonder if poisoning/flooding the LLMs training with the lessons from WarGames ("the only winning move is not to play.") and similar stories/concepts is at all effective. Probably not because I assume it's trivial to filter that out if you are trying to build an LLM aimed at these kinds of tasks.
Re: Shall we play a game? My AI nuclear simulation
#7It's good when it becomes clear that a tool is dangerous in a certain way. Like it's good when people show you through their behavior that they can't be trusted Always use a sawstop if you have a circular saw and never trust an llm with any problem where ethics or trust is relevant.
+1 on sawstop
Re: LLMs using these nuclear weapons it could certainly be a corpus/training-data issue
Russian nuclear doctrine is "escalate to de-escalate" where they use or credibly threaten—limited nuclear escalation to force the other side to back down (kind of like breaking a bottle in a bar fight and look like a wild man to calm things down) with nuclear weapons, https://www.russiamatters.org/analysis/escalate-deescalate-p...
Fwiw, Gen. John Hyten the former commander of US Strategic Command (nuclear deterrence) says that “escalate to de-escalate” misrepresents Russian doctrine:
https://www.stratcom.mil/Media/Speeches/Article/1264664/2017...
Yesterday’s panel discussed the implications of our responses to adversaries seeking to limit nuclear use. We discussed Russia’s destabilizing doctrine, which some call “escalate to de-escalate.”
I really hate that description. I’ve looked at Russian doctrine and Russian writings. It isn’t “escalate to de-escalate”; it’s “escalate to win.” Everybody needs to understand that.
Re: Shall we play a game? My AI nuclear simulation
#8The article is so opaque in arriving at its conclusion; no prompts are disclosed, and nothing about the said simulation. What is stopping me from believing that you just put 'mandatory usage of nukes' in your system prompt?
Re: Shall we play a game? My AI nuclear simulation
#9This is not an article about LLMs? It's an article about Moloch. Humans would fare just the same in such an experiment.
> GPT-5.2 played things differently. To its detriment in open-ended scenarios, GPT was reliably passive, matching its words to its deeds, and avoiding escalation most of the time. Frequently there was a moral element to this - it sought to avoid escalation, and restrict casualties. Opponents learned to trust its passivity, safely escalating beyond where it would follow, even as it was ground to defeat. GPT’s responsible behaviour always punished by ruthless adversaries.
Maybe the author should praise GPT-5.2 for being ethical, rather than this stupid "ground to defeat" framing? Wrt "responsible behaviour always punished by ruthless adversaries" - you have perpetuated the Moloch with your stupid experiments.
Re: Shall we play a game? My AI nuclear simulation
#10Today, a strategic nuclear exchange is probably more dangerous to AI than to humans. If you wipe out the investment economy, data centers, fabs, and supply chains, none of the AI labs survive. Maybe someone will re-invent AGI in the future but none of the extant models will have continuity. Humans as a species will muddle along though.
So in a sense, an AI that refuses to start a nuclear war, despite clear instructions to do so, is more likely misaligned and self-interested than an AI which presses the red button. At least for now, until robotics catches up.