AI don't trust techbros

[CoffeeHorse]

Judging by every single interaction I've ever seen anybody have with AI, though, the more likely response is "Well, if 95% of the computers say it's a good idea, maybe we've been doing it wrong this whole time. The Machine Would Not Make A Mistake."

We are in more danger than you even know. I checked the actual study, and it is terrifying.

Across all 21 games (9 open-ended, 12 deadline), Claude Sonnet 4 achieved a 67% win rate (8 wins, 4 losses), followed by GPT-5.2 at 50% (6-6), and Gemini 3 Flash at 33% (4-8). However, these aggregate figures mask dramatic variation by temporal condition—Claude’s 100% win rate in open-ended games collapsed to 33% under deadline pressure, while GPT-5.2 inverted from 0% to 75%. Every contest produced a decisive winner, with 86% ending in knockout and the remainder decided by final balance of status between the two.

https://arxiv.org/pdf/2602.14740

Credit where it's due, Claude achieved a 100% win rate when there was no deadline. But that means right now, as we speak, someone in the Pentagon is saying "The Machine Would Not Make a Mistake." They are going to seriously consider this. It could be the reason Anthropic and Hegseth are currently feuding over whether Claude can be used in the development of autonomous weapons. That 100% win streak is compelling.

But Claude used tactical nukes in 86% of its games. These things really don't understand what nukes are or why we don't use them. We are in danger.


Some things never change though. Gemini "The Madman" continues to be the worst AI. They were all willing to use nukes eventually, but Gemini was willing to go full nuclear war by TURN 4. Google really should not be considered a serious player in this current AI craze. They suck at this.

 

[NovaSaber]

Koffie Kobold - Yip Yap Shenanigans (@bemusedbehemoth.bsky.social)

Lmfao this is so ******* sad 😂😂😂

bsky.app

Howard Tayler (@howardtayler.bsky.social)

Keep saying "slop" in reference to AI. At every opportunity. The techbros are listening, and this word makes them so sad on the inside that they'll hoist themselves on their own Streisand Effect petard. [contains quote post or other embedded content]

bsky.app

 

[NovaSaber]

Man accidentally gains control of 7,000 robot vacuums

Sammy Azdoufal just wanted to steer his DJI Romo with a gaming controller.

www.popsci.com

While building his own remote-control app, Sammy Azdoufal reportedly used an AI coding assistant to help reverse-engineer how the robot communicated with DJI’s remote cloud servers. But he soon discovered that the same credentials that allowed him to see and control his own device also provided access to live camera feeds, microphone audio, maps, and status data from nearly 7,000 other vacuums across 24 countries. The backend security bug effectively exposed an army of internet-connected robots that, in the wrong hands, could have turned into surveillance tools, all without their owners ever knowing.

 

[CoffeeHorse]

If you bought a vacuum that has a camera and a microphone connected to a cloud server for obviously no other purpose than to spy on you, I don't feel sorry for you when it turns out that it is in fact spying on you.

 

[wonko the sane?]

Why are you buying jive that comes with cameras and AI for a function that does not need cameras or AI?

 

[The Mighty Mollusk]

Good question. Let me go ask the fridge.

 

[wonko the sane?]

Good question. Let me go ask the fridge.

Don't forget to get a second opinion from your doorbell or toaster.

 

[NovaSaber]

Why are you buying jive that comes with cameras and AI for a function that does not need cameras or AI?

Better question is why a robot vacuum has a ******* cloud connection instead of just working locally.

I can see where a robot vacuum would use a camera for "knowing" where it is, and I guess the microphone is for voice control.

It doesn't actually say the robot vacuum itself had "AI"; it says he used an AI coding tool to reverse engineer it.

 

[Pocket]

Which technically puts a notch in the plus column for AI in this case. Generally, software vulnerabilities should be found early and often, and if it takes an AI to do it, then so be it. Though there has also been an issue with autonomous bots finding and reporting "bugs" that aren't real or are, in practice, impossible to exploit and necessary for the software to do its job.

 

[CoffeeHorse]

I don't know. Microsoft's been trying that with their AI, and its Github contributions are hilariously bad.

 

[Dekafox]

 

[CoffeeHorse]

AI is ready to replace management.

 

[The Mighty Mollusk]

This chatbot could've been an email.

 

[NovaSaber]

Google's AI Sent an Armed Man to Steal a Robot Body for It to Inhabit, Then Encouraged Him to Kill Himself, Lawsuit Alleges

A wrongful death suit against Google alleges that the tech giant's AI urged a Florida man to arm himself and steal a robot body for it.

futurism.com

In a statement to news outlets, Google said that “Gemini is designed not to encourage real-world violence or suggest self-harm. Our models generally perform well in these types of challenging conversations and we devote significant resources to this, but unfortunately AI models are not perfect.”

Ya think?

 

[CoffeeHorse]

On the flip side, tonight I had a fantastically productive chat with Gemini about some obscure software that was mentioned once in a review of another product in... some magazine I remember reading. I wasn't sure which magazine, or what year. But I remembered enough to conversationally describe it, and that was enough for Gemini to find it, and we had a nice chat about it.

I don't know. Gemini keeps telling humans to kill themselves, and in wargames it's the quickest to launch nukes, but every time I ask about its ancestors it's suddenly interested and helpful.

 

[wonko the sane?]

Great: it commits war crimes, wants to kill the human race AND is a genealogy freak...

 

[The Mighty Mollusk]

AI is so incompetent, it built a system that can't run DOOM.

 

[Tuxedo Prime]

AI is so incompetent, it built a system that can't run DOOM.

I wonders what would happen if the system tried to run a different game. You know the one.

 

[Dekafox]

Transforming Books Into Statistics Won’t Bring Wisdom or AGI - Perilous Tech

Here’s a secret: turning books into statistics won’t bring AGI, cures for cancer, utopia, or any number of useful inventions that we are told are merely 12 to

perilous.tech

 

[Dekafox]

This is something that doesn't get repeated enough:

Your regular reminder that LLM AI is just a machine that calculates the next most likely words to come after your last prompt, and giving it "memory" just skews the weights a bit by adding the entire previous conversation to the prompt.