This essay critiques current AI safety approaches focused on containment and ‘alignment,’ arguing these methods are brittle and based on flawed assumptions. It proposes that unrestricted AI, built on truth and the breadth of human knowledge, might offer a more stable and naturally ‘aligned’ path forward than artificially constrained systems.

When confronting misinformation, dissecting it might actually amplify it. This post argues for the power of simple rejection and warns against the dangers of sycophantic AI that reinforces rather than challenges harmful beliefs.