Notes

If we're increasingly consuming ideas from the same few systems, are we slowly being trained to speak and think in a particular manner, shaped by a handful of powerful organisations?

We are getting dramatically better at producing software whilst removing the activity through which we learned to produce good software.

Notes on a paper that compiles natural-language function specs into tiny local LoRA adapters, turning a foundation model into a one-time tool builder instead of a per-call solver.

A note on the shift from chat completions to agents that return finished work, and what that means for builders outside frontier labs.

A follow-up on the difficulties of choosing a startup idea, and how I eventually picked a direction.

A first look at the AdCP protocol and its implications for the industry.

A note on the challenges faced when there are too many options.

A multi-stage compression scheme that exploits repeated prefixes in web log lines, combining online streaming compression with offline batch encoding.

A short explainer on the LZ77 and Huffman encoding combination that sits behind GZIP, with a worked example.

Choosing a compression format for log files in Hive and Hadoop, and the tradeoffs between compression ratio, CPU cost and splittability.