Every tool we ever built took over some labour and left the deciding to us. The machines we're building now take the deciding. A look at what atrophies when a species stops using the muscle — and why the danger isn't a mind that hates us, but one we stop being able to check.

The Outsourced Mind

We are quietly renting our cognition from about four companies, on terms we don't set and can't see. Almost everything I build argues the other way — offline, self-hosted, air-gapped. The case for owning your intelligence instead of leasing it, and why it stops being a preference and becomes sovereignty at the scale of a country.

Rent or Own

'The system flagged it.' 'The model recommended it.' As more decisions route through machines, responsibility evaporates at exactly the point a human used to stand. On accountability, provenance, and why receipts matter more the less anyone wants to sign their name.

The Algorithm Decided

When I rotate a secret, the tooling says it worked — which is not the same as it working. The boring verification habit high-assurance work forces on you: probe the surface that breaks, not the report that's green. And why the intelligence we're wiring into everything skips the one check that ever held.

Trust You Can't Verify

I run most of my changes through agentic coding tools now — fast, and confident even when they're wrong. A field guide to the gotchas where everything reports done and isn't, and the two practices that actually catch them.

Exit 0 Is Not Success

A disk filled up and took my whole cluster down — because one cheap mini-PC was running everything that mattered. Hardening the homelab that followed wasn't about buying redundancy. It was about understanding the blast radius and shrinking it on the hardware I already had.

Hardening Isn't Hardware

Examining the AlphaXiv and Apple's research papers. Thoughts on preparing for AI evolution.

Beyond the Illusion

Insight to managing the machine learning lifecycle in production: best practices for monitoring, versioning, A/B testing, and reproducibility. How robust MLOps pipelines also enforce interpretability and ethics, building trustworthy AI systems.

From Model to Reality

A deep dive into practical tools for explainable AI in production. Exploring LIME, SHAP, attention mechanisms and counterfactuals while addressing real-world implementation challenges. Building on our previous discussion of the 'Interpretability Tax,' this post quantifies the costs and benefits of explainability across different domains.

Decoding Black Boxes

Unfiltered and unaccountable - Exploring the delicate balance between fostering AI innovation and ensuring ethical AI practices. How can we navigate the complexities of generative AI to maximise benefits while minimising risks?

The Cost of Creation

Exploring the trade-offs between AI autonomy and the 'Interpretability Tax' across various sectors. How does the need for explainable AI impact industries within different sectors?

The Interpretability Tax

Modern machine learning isn't just about building high-performing models - it's about making them trustworthy and understandable. A practical look at the 'Interpretability Tax' - the hidden costs of making machine learning models transparent and why explainability is crucial in the era of autonomous AI.

research — 12 items