Free AI For Everyone, Says Everyone; Nobody Disagrees Out Loud
Scott Alexander weighs the debate over open-weights AI, where companies like Microsoft, NVIDIA, OpenAI, and Meta have signed a letter supporting free, downloadable AI models. The main concerns are criminal misuse (hacking, bioterrorism) and the impossibility of aligning models that anyone can retrain. Alexander remains neutral, arguing that superintelligence takeover risk warrants preemptive action but criminal misuse risk does not—society will overreact after the first incident, as it always does, and that will be soon enough.