Research-oriented articles on symbolic systems, model tooling, and technical architecture.
Article
Go gives you errors as values and, since 1.21, structured logging in the standard library. The two do not meet: context is known deep in the call stack, and the decision to write a log record belongs at the boundary. A small type closes ...
Article
Claude Fable 5 was public for three days before an export-control letter made it unreachable. That week settled four arguments I'd been having since January, and exposed the borrowed floor that holds the price of everything else down.
Article
Thoughts on collaboration, authorship, and the tools that extend our capacity to think and write
Article
Vulnerability discovery is becoming a function of targeted AI compute, turning software security into a fire hose problem of spend, access, and remediation capacity.
Article
A toolchain for testing whether the Exp-Minus-Log representation gives symbolic regression a useful search space.
Article
Alphabet's AI problem is structural, not mainly competitive. If AI systems weaken Google's control over intent capture, they threaten the advertising engine that funds the company while the time left for a serious response keeps shrinking.
Article
As frontier models get stronger, access may become more stratified. The token divide is a market structure in which the best models, highest limits, and most productive agent workflows are divided by price.
Article
Anthropic's Mythos may be too dangerous for public release. Project Glasswing raises the harder question: when defensive access is scarce, who gets protected first, and who waits outside the perimeter?
Article
Google’s Gemma 4 release does more than expand an ecosystem. It changes which AI companies are exposed when model access becomes less scarce.
