Pentagon used Elon Musk’s Grok AI to fire missiles at Iran, official says
www.independent.co.uk
Top defense official says data centers powering trillionaire’s chatbot are critical to national security

Donald Trump’s administration turned to Elon Musk’s Grok chatbot to launch thousands of missiles in Iran, according to a top defense official.

In a sworn statement defending the trillionaire from a lawsuit alleging xAI data centers are illegally polluting Black communities, the Pentagon’s artificial intelligence chief said the chatbot’s continued operation is “a matter of paramount national security” — and was used to fire more than “2,000 munitions at 2,000 distinct targets within 96 hours.”

Grok, a generative artificial intelligence chatbot developed by xAI, is among four AI models “currently capable of supporting national security applications,” according to Cameron Stanley, the Pentagon’s chief digital and artificial intelligence officer.

  • NuXCOM_90Percent@lemmy.zip
    3·
    3 days ago

    Okay, but what does that mean? How does an llm do that exactly? What are the inputs and outputs?

    Not an LLM. An inference model (which… we group in with LLMs these days because stupid).

    Let’s keep it simple and say you have access to a bunch of traffic cameras. You are on the lookout for… I dunno, let’s say a handsome young italian american in a dark green coat with a face mask on.

    1. Run a computer vision tool to identify objects in all of the camera feeds. This is trivially parallelized and often done on the device/hub itself (“edge computing”)
    2. Put every feed that triggered “homeboy” into a bucket
    3. Convert that into a more easily worked on bit of data. A CSV of Timestamp, location, and confidence tend to work well
    4. Pass THAT into a different inference model to detect patterns. For example, we had a pip at 4 pm on 5th street and another one at 4:05 pm on 6th street. Make that a possible path that was taken.
    5. Homeboy has maybe six different possible routes he took through the city, and three of them end with the appearance of “handsome young italian american in a dark green coat without a face mask”
    6. One more pass with a different model to collate that with training data (“We tracked Fred walking between cameras”) and you now have maybe two probable options, each with their own weight as to likelihood of being the completely innocent young man.

    Just apply that on a larger scale using satelite surveillance data, intel on known targets, etc.

    Gonna actively ignore all of your obnoxious ad hominem in the attempt to have an actual conversation. Don’t prove me the fool for doing so

    But we’re specifically talking about grok here. How does grok fit into the process of firing a missile exactly, and what are the incentives built into the system for the people involved?

    Which gets us to LLMs. LLMs… are REALLY good at translating from one language to another. For most people, that is natural language (a prompt) to machine instructions that get fed to underlying software (“agentic models” if you want to use buzz words). For ChatGPT that is usually just a google search. Or it is a few bullet points that turn into a letter to Grandma. Or a question that is turned into a non-answer.

    And then… the rest is exactly what I said. “Find me all instances of General Whatshisface” is a query that would map to a database lookup that would use the already labeled (either through computer vision or a human analyst) data to find General Whatshisface. And then it is exactly what I said above

    The idea is that “AI” is used to detect possible targets and “write the first draft” of the report suggesting a strike based upon available intel. It is then used to flesh out that “draft” into a full report. It is THEN used to determine which reports are valid. And THEN it is used to draft the orders to whoever is firing the missile.