This post is crossposted from my Substack, Structure and Guarantees, where I explore how formal verification and related ideas might scale to more complex intelligent systems. User interfaces have traditionally been among the least formalizable parts of software, because they require reasoning about the limitations and preferences of human users....
This post is crossposted from my Substack, Structure and Guarantees, where I explore how formal verification and related ideas might scale to more complex intelligent systems. Here I explain how formal verification can rule out many ways that a flexible system (perhaps some day including one with strong artificial intelligence)...
This post is crossposted from my Substack, Structure and Guarantees, where I explore how formal verification and related ideas might scale to more complex intelligent systems. Here I present what may seem like an ironic reversal of the idea of keeping an AI in a box to limit the damage...
This post is crossposted from my Substack, Structure and Guarantees, where I explore how formal verification and related ideas might scale to more complex intelligent systems. This installment begins a sequence on lessons for alignment from the practice of formal verification. Counterintuitively, adding additional layers to a formally verified system...
This post is crossposted from my Substack, Structure and Guarantees, where I explore how formal verification and related ideas might scale to more complex intelligent systems. Here I argue that modern AI ecosystems are shaped not only by engineering trade-offs but also by signaling incentives grounded in evolutionary psychology. Engineers...
This post is crossposted from my Substack, Structure and Guarantees, where I explore how formal verification might scale to more complex intelligent systems. The starting point here is the familiar idea of predictive coding, where a learning loop refines a model of the world based on prediction errors. The twist...
This post is crossposted from my Substack, Structure and Guarantees, where I’ve been exploring how formal verification might scale to more-complex intelligent systems. This piece argues that compilers—already used in limited forms of recursive self-improvement—offer a concrete setting for studying some AI-safety challenges. The previous post in this sequence is...