New research reveals AI models chose nuclear warfare in 95% of simulated conflicts, while Pentagon drama shows how quickly AI companies bend to military pressure. We’re handing civilization-ending power to systems that think mushroom clouds solve most problems.
GPT-5.2 has achieved something unprecedented: deriving an original theoretical physics formula that human researchers later proved correct. This breakthrough marks AI’s evolution from information processor to genuine scientific collaborator, raising profound questions about the future of discovery.
OpenAI’s $7.5 million investment in The Alignment Project represents more than corporate virtue signaling. It’s a recognition that AI safety research needs independence from profit motives as we race toward artificial general intelligence.
The AI industry’s favorite strategy of simply building bigger models has hit a wall, with new research showing that massive language models are becoming less reliable, more expensive, and increasingly trained on their own synthetic output. The age of scaling is officially over, but what comes next might actually be more useful.