Cool, so if you worked on Grok at X, there's a decent chance you materially contributed to the death of 120 schoolgirls in Minab. Great job guys. I hope you're proud of the cause you chose to contribute to.
Of course, that might not stand out to the average American given the rate of violent attacks on our own schools.
Yeah as soon as the early reporting came out about the use of LLMs during the operation, it kind of made sense thinking through how a target like that could be selected. The LLM is going to use variations of clustering algorithms, calculated centrality scores for all the targets, and tried to find a location where they were all in the same place at the same time (for bombing purposes)... It just so happens there are only a few schools that the Iranian elite have their children attend. And without the context of a human seeing the target and knowing it's a school, well, that's how you get this result.
At least in the Terminator, etc, there is debate about building skynet and giving it control... In reality it's just like yeah go ahead let it loose.
The article is valid alarm but I would expect more education from "The Hill" to know that it wasn't the chatbot version of Grok, saying that makes the author sound stupid
I could see how the government’s “equivalent to 1.5 billion words or 6 million pages of text being processed by the technology” statement could be interpreted as chat.
Cool, so if you worked on Grok at X, there's a decent chance you materially contributed to the death of 120 schoolgirls in Minab. Great job guys. I hope you're proud of the cause you chose to contribute to.
Of course, that might not stand out to the average American given the rate of violent attacks on our own schools.
- https://en.wikipedia.org/wiki/2026_Minab_school_attack
there's a non-zero chance Grok accidentally bombs Lebanon, Oregon or Moscow, Idaho in the future
Now it makes sense why they hit a girls' school instead of an actual military compound.
Yeah as soon as the early reporting came out about the use of LLMs during the operation, it kind of made sense thinking through how a target like that could be selected. The LLM is going to use variations of clustering algorithms, calculated centrality scores for all the targets, and tried to find a location where they were all in the same place at the same time (for bombing purposes)... It just so happens there are only a few schools that the Iranian elite have their children attend. And without the context of a human seeing the target and knowing it's a school, well, that's how you get this result.
At least in the Terminator, etc, there is debate about building skynet and giving it control... In reality it's just like yeah go ahead let it loose.
The article is valid alarm but I would expect more education from "The Hill" to know that it wasn't the chatbot version of Grok, saying that makes the author sound stupid
"The concern raised is valid and should cause alarm"
BUT
"I'm going to ignore it instead because of a disagreement on the terminology used"
I could see how the government’s “equivalent to 1.5 billion words or 6 million pages of text being processed by the technology” statement could be interpreted as chat.
ok, so this comment is just _as_ stupid given there's only two AIs of note being used: image models and LLMs.
Neither of them should be trusted with decision making.