• 16 Posts
  • 118 Comments
Joined 1 year ago
cake
Cake day: June 16th, 2023

help-circle






  • No, but the US military has never had their homeland logistics fucked with in recent history. Sure you can’t easily destroy a Bradley APC, but it needs fuel that happens to be stored and transported in ways that are not as resistant to attack. And when the fuel runs out many vehicles are no longer useful in combat.

    Or spare parts. Germany got their industries bombed like crazy in WW2. Even though their stuff was better on paper they didn’t have the parts to keep combat effective. Ask any veteran how reliable military vehicles are without constant maintenance.

    This is hypothetical and all, but it’s not that big of a stretch of the imagination to see any American insurgency becoming a real pain in the ass for the military over months and years. And unlike Afghanistan they can’t simply withdraw when they’ve had enough.










  • The heads up display isn’t something you see in front of you like most planes. The helmets are the heads up display, like augmented reality.

    There are cameras all over the plane to help you see through the aircraft (see ground targets through the floor, nearby aircraft through your wing). Think of the resolution and bitrate needed to make it useful!

    Just like how an apache gunner can simply look at a target to aim the gun at them you can do the same thing to get missile lock. And if you can’t hit it it’s still marked for every allied plane in the airspace to see. If you are out of missiles but you are tracking an enemy plane miles ahead, you can send the data to an F-15 miles behind you and let their missiles lock and fire from farther than they can engage alone.

    With that in mind the radar is awesome letting it see threats from greater distances than the opposition, with the stealth capabilities good enough to keep them from easily doing the same.

    I’m sure there are other surprises too, but the military obviously wants to keep those a secret





  • I genuinely think the best practical use of AI, especially language models is malicious manipulation. Propaganda/advertising bots. There’s a joke that reddit is mostly bots. I know there’s some countermeasures to sniff them out but think about it.

    I’ll keep reddit as the example because I know it best. Comments are simple puns, one liner jokes, or flawed/edgy opinions. But people also go to reddit for advice/recommendations that you can’t really get elsewhere.

    Using an LLM AI I could in theory make tons of convincing recommendations. I get payed by a corporation or state entity to convince lurkers to choose brand A over brand B, to support or disown a political stance or to make it seem like tons of people support it when really few do.

    And if it’s factually incorrect so what? It was just some kind stranger™ on the internet