US Army tests OpenAI models in simulated war games
The US Army Research Laboratory is testing OpenAI's GPT-4 Turbo and Vision models to assist with operational planning inside military video games. While GPT-4 outperformed older AI models in formulating attack and defense strategies, researchers found it also caused significantly more casualties to achieve mission objectives. Why it matters: as militaries accelerate AI adoption, models that prioritize ruthless efficiency highlight mounting ethical and safety risks.