So Skynet is real and the movie War Games was a documentary?

[petergleick@fediscience.org]

So Skynet is real and the movie War Games was a documentary?

AIs can’t stop recommending nuclear strikes in war game simulations

Leading AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases

New Scientist (www.newscientist.com)

[distractal@hachyderm.io]

@petergleick At least in War Games the AI was smart enough to realize that in every scenario nuclear war means total loss for all sides.

These singularly idiotic motherfuckers want to connect the "glue is good on pizza" machines to military systems.

[lcwheeler@ecoevo.social]

@petergleick What percentage of the LLM training data surrounding nuclear war is made up by the Terminator franchise and War Games? So depressing. SciFi authors always envision these grandiose stories about hyper-intelligent AIs turning on humanity, but the reality is SO MUCH DUMBER. Feels more like the kind of tech apocalypse from Cat's Cradle than the one from Terminator.

[ghostonthehalfshell@masto.ai]

@petergleick

The greatest threat to humanity is not super intelligence, but the machine mindlessness of the men who own the machines and the enormous power that they have captured

[akamran@indieweb.social]

@petergleick @samiamsam life imitating art

[jmcs@social.jsantos.eu]

@lcwheeler @petergleick if they trained it on Wargames it wouldn't be this stupid.

[cliffsesport@mastodon.social]

@distractal @petergleick gah I don't grok anyone that really tests these word shufflers would trust them for cooking. It doesn't take much testing w Gemini to realize it often won't adjust even when you point out errors.

[rrb@infosec.exchange]

@petergleick Assuming that the data center is well shielded against the EMP, this is a very rational choice. Von Neumann would be proud.

[rrb@infosec.exchange]

@bencourtice @petergleick I am starting to hope for thermo-nuclear obliteration just to stop the stupidity. I pray that the cockroaches learn from our failures.

[http_error_418@hachyderm.io]

@petergleick the writer still falls in to the trap of calling the output "reasoning", I didn't even need to get past the paywall to see that. Not worth my time to look further, that's enough to see that it's bad reporting.

[simplicator@federate.social]

@petergleick Musk named his “AI” supercomputer “Colossus” https://en.wikipedia.org/wiki/Colossus%3A_The_Forbin_Project

[simplicator@federate.social]

@lcwheeler @petergleick & sci-fi writers & Hollywood make it seem super-cool
See also the movies “Wall St.”, “Wolf of Wall St.,” etc.

[henstridge@mstdn.social]

@petergleick

This should fix it:

XOO
OXX
XXO

[infrapink@mastodon.ie]

@petergleick
No, because the computer in War Games had yo be goaded into using nukes.

[vgoller@nrw.social]

@petergleick won’t hurt AI. Just saying

[xro@chaos.social]

@petergleick
Real humans, at some point, do the total cost calculation of afterwards having to live in constant fear of having the same done to them as well as the damage to humanity, international relations and trade and their own standards of living...

And decide that this is not worth it.

#AI just gets told "to win" and executes without the human subtext

[avincentinspace@furry.engineer]

@petergleick just in case this wasn't scary enough: https://apnews.com/article/anthropic-hegseth-ai-pentagon-military-3d86c9296fe953ec0591fcde6a613aba

[tarheel@mstdn.io]

@xro @petergleick @cstross

The real piece of info here would be: how much of this total-cost calculation must be included in the wargame to make the AI choose another path, and is THAT cost realistic? Because if not... duck and cover.

[yojimbo@masto.hackers.town]

@petergleick Ah, look, there's the mistake :-
"The AI models ... produced around 780,000 words describing the reasoning behind their decisions."
No, they produced 780,000 words that looked statistically similar to what a human might have said under the same circumstances.

[gelsomore645@social.vivaldi.net]

@petergleick