Ces's random twalks on tech and stats.
In May I argued AI sounds like AI because it writes like the average of us. Four pre-registered studies found the tells are ours, with an open model using some far more than any human writers do.
(September 29, 2026)
Warnings about disaster do the most good when they say how it would happen and what to do about it. What economic history, warning research, forecasting and AI safety know about that.
(September 18, 2026)
If AI ends up writing near-perfect code, do software updates become new features only? No. Some change you choose and some the world forces on you, and AI may soon handle the forced kind on its own.
(September 18, 2026)
Send an agent the source. In my tests a web link gave it a short summary, an artifact link failed or cost up to twice the page, a Quarto page defeated its Read tool, and PDFs broke file paths.
(September 17, 2026)
A tokenwise route has a cost of its own. What asking costs from three Claude Code settings, and the task size below which the answer costs more than it saves.
(September 11, 2026)
On a subscription the cost of a task is a share of your usage limit. A Claude Code plugin for not spending that share on work that never needed it.
(September 9, 2026)
I banded monthly downloads by powers of ten without checking where the packages actually were. A sample of 600 put 532 of them in one band, and the band below caught nothing at all.
(September 4, 2026)
Adding a dependency takes seconds and removing one can take weeks. CRAN already publishes enough to answer the question before you commit — it’s just spread across three APIs most of us don’t check.
(August 22, 2026)
Every R modelling package returns results in a different shape, so reporting code breaks whenever the model changes. What one tidy interface across twenty algorithms buys you.
(August 21, 2026)
The cloud has come apart into components. When a production stack needs no hyperscaler console, the question stops being whether you can run it and becomes which option to choose.
(June 2, 2026)
The em-dash, the tripled list, the careful hedging: the tells people read as machine fingerprints were learned from us. On whose writing gets smoothed out, and who gets accused.
(May 4, 2026)
AI is held to a standard humans are not, and the usual defence is that fallibility comes with human-likeness. That premise does not survive contact with how we already judge each other.
(May 4, 2026)
An AI code review returned 27 findings. Asking how many were left to find turned out to be the more useful question, and the harder one to answer.
(May 4, 2026)
A surgeon asked how a phone does what a far bigger computer does. It mostly doesn’t; it’s a frontend. Computing keeps swinging between central and local, and the swing has a pattern.
(March 20, 2026)
Rebuilding a developer machine is dozens of tools, hundreds of packages and the one environment variable you always forget. I handed the migration to Claude Code instead.
(February 11, 2026)
A follow-up on proprietary languages: if AI can write in any of them, what actually decides which languages survive, and whether we will still need to read them.
(December 12, 2025)
Open source won on trust and access long before AI. What AI changes is the last argument for paying: the cost of knowing the language.
(September 11, 2025)