Latest Posts

The Slipbox maintenance loop on video, and the protocol I wrote to make the same vault double as an AI agent's memory across sessions.

The LLM you're coding with is a product, tuned by people whose business depends on your next prompt. That incentive quietly punishes an honest "I'm not sure," because it reads as weakness, and it pays out for a confident answer that sounds right whether or not it actually is.

slipbox-mcp gives my notes typed links and hands the agent the tending and upkeep I was never going to keep up with myself.

Worktrees split your work across directories, and after enough context switches you stop keeping track of which is which. So I wrote eleven lines of shell to stop tracking it.

Turing proved in 1936 that universal verification is impossible. Now we're trying it anyway, on AI systems that adapt to whatever detection we point at them.

Hand me a detector f and I can build a program g that defeats it. The same trap catches alignment testing: every test you run is one more signal telling the model humans are watching.