Essays about software, machine learning, research, and the occasional detour into everything around them.
Not AGI yet, but clearly in its orbit.
GPT-6 Astra can operate computers, write tools, navigate novel environments, and carry context across long tasks. It is not AGI, but it may be our clearest picture yet of almost-AGI.
Anthropic is moving AI from “please summarize this” to “please calibrate the laser, carefully”
Anthropic's MCP connected AI agents to software and data. MHS is an attempt to give them a safer, more standard way to operate physical hardware.
A beginner-friendly tour of Kubernetes, kubectl, and why AI infrastructure gets real after the demo works.
Kubernetes helps keep containerized applications running across machines. This is the simple explanation I wish I had when I started learning it, plus the kubectl commands that are actually useful.
How Transformers became modern AI's default architecture, and why quantum is still looking for its own scalable abstraction/convergence
Transformers won because they matched model design, hardware, and economics at the same time. Quantum computing becomes interesting when AI's next bottlenecks stop being purely algorithmic and start becoming physical.
Mechanistic interpretability, AI safety, and my first blog
Mechanistic interpretability is the field of asking what actually happens inside the model, and why that matters once the systems stop being toys and start becoming infrastructure.