A standing habit around here: when we read something clear-eyed about putting AI to real, public-minded work, we keep it. No hype, just things worth your time.
01 · Introducing Claude Opus 4.8
The newest model from Anthropic. The number that caught our eye wasn’t a benchmark, it’s that Opus 4.8 is reportedly about four times less likely to let a flaw in its own code slip by unremarked, and more willing to say when it’s unsure. For high-stakes work, a model that flags its own doubts is worth more than one that’s merely confident.
02 · A 10-year-old Xeon is all you need
Christina Sørensen runs a capable 25-billion-parameter open model on a 2016 server with no GPU. It’s more proof we’ve seen that the barrier to private, on-premises AI is understanding the machine, not buying a data center.
03 · AI reprices public-sector knowledge work
A sharp May 2026 read on what AI is doing to the economics of government knowledge work. TL;DR: the cost of routine drafting and review is collapsing, and the agencies that come out ahead won’t be the ones buying the most AI, but the ones that treat governance as operating infrastructure.
04 · Dartmouth Health on AI in rural healthcare
Physicians making the honest case: AI can widen access for rural patients, but only alongside broadband and real infrastructure. That tension, the promise and the plumbing, is exactly what public-health work in Alaska runs into.
05 · Machine learning for electricity load forecasting
A readable tour of how utilities use machine learning to forecast electricity demand and hold the grid steady as wind and solar make it less predictable. From 2025 and still relevant.
Reading something we should be? Send it our way through the contact form. And if you’d like a note when the next list lands, the subscribe form on any page does the trick.
— Tyler Arnold
All posts