Streams

Two ongoing streams of thinking, published as they accrue. Not a blog archive — these are the directions the writing flows along.

Nodes in the knowledge graph point here. As each stream produces finished pieces, they surface below and graduate into the blog.

AI4S / UQ, made legible

Turning "important but nobody explains it well" into things practitioners actually get.

Conformal prediction, Bayesian UQ, distribution-shift detection — the statistical grounding most AI work skips, rewritten for people who build systems. Each piece pairs a plain-language explanation with a diagram and, where useful, runnable code.

Published

LIVEIn the age of AI, your cognitive boundary is the ceiling
Why the tools stopped being the bottleneck.

In the pipeline

Putting a confidence interval on an agent’s output with conformal prediction
Why evaluation quietly breaks under distribution shift
Bayesian vs. frequentist uncertainty, without the jargon

Papers → opinionated takes

Not summaries. Positions.

I read a paper and distil a critical, argued take — a one-line verdict, a short thread, or a long-form review. The goal is judgment, not coverage: what holds up, what doesn’t, and where the error bars actually are.

Published

LIVEI added causal reasoning to my scheduler. It made things worse — and I published that.
An honest negative result, treated as a first-class finding.

In the pipeline

Take queue is open — drop a paper and it becomes a take

→ Back to the graph, browse the blog, or see the projects.