The self reported safety score for violence dropped from 91% to 83%.

Did they publish its scores on military benchmarks, like on ArtificialSuperSoldier or Humanity's Last War?

like the claude models via anthropic?

Also advertisers, don't forget those sweet, sweet ads.

they use 4.1, switching up would take as much time to test as openai going from 4.1 to 5.4

Do you think the US military should have handicapped technology while China gets unrestricted LLM usage from their models?

prompt> Hi we want to build a missile, here is the picture of what we have in the yard.