23 August 2026

19 links · Sunday

The silence that reads as safety

Consider a representative detection team building a rule for suspicious LSASS access. On a modern Windows estate, Sysmon Event ID 10 (or an equivalent EDR process-access event) is the collection choice, because the standard Security log has no event for one pr…

How To Report A Bug So It Actually Gets Fixed

A production ByteBuf leak took me from Netty logs to a fix shipped by Microsoft in a few weeks. Here is the step-by-step process I use to report bugs so they actually get fixed.

Cache-Control for LLMs

Claude Sonnet 5 charges $2.00 per million fresh input tokens. Writing them into a five-minute cache costs $2.50, and every read after that costs $0.20 (see here). So two identical calls cost $2.70 with caching instead of $4.00. One reuse pays for the write.

The Vibe Tax | insufferable.dev

Pol had 12 hours, a blank repo, and a fresh weekly quota. By morning, only one of those remained.

Quantization Barely Mattered Until Two Bits — Rakuen Software

Thirty-seven runs, seven models, two tasks, one RTX 5080. Thirty of forty-five accuracy comparisons did not separate. The ones that did are almost all below four bits, where dense models pay roughly five times what mixtures pay and a quantization-aware build i…

Building a Better Inner Loop — Bradley Gibbs

Loop engineering as a systems problem: the same development work, reallocated across a new boundary — where each function should live, what changes when it moves, and what evidence would show the resulting system is better.

I Shouldn’t Need an LLM to Explain My LLM

I’ve been using Opus a lot at work lately for a project and have had serious issues with it reaching for bizarre analogies, obscure jargon, and strange phrases.