Blog
Posts tagged #artificial-intelligence
of 11 - 11 NUL bytes in generated code: source files that compile, lint, and run — but are invisible to grep, VS Code search, and code review A raw NUL byte inside a string literal leaves a TypeScript file valid to tsc, node, ESLint and Prettier, and invisible to ripgrep, VS Code search, and git diff — measured on each. The key-format bug that hid behind it, how a JSON tool call writes the byte, and the two fixes that need no vigilance. Aug 22, 2026 · 10 min
SELECT country, region, hex(region) FROM wa_geo_daily WHERE region LIKE '%exas%';
-- US | Texas | 5465786173
SELECT country, region, hex(region) FROM wa_events WHERE region LIKE '%exas%';
-- US | Texas | 5465786173 SELECT country, region, hex(region) FROM wa_geo_daily WHERE region LIKE '%exas%';
-- US | Texas | 5465786173
SELECT country, region, hex(region) FROM wa_events WHERE region LIKE '%exas%';
-- US | Texas | 5465786173 - 08 How far synthesized speech overruns subtitle timing: 150 measured cues A measured distribution of the compression synthesized speech needs to fit real caption timing, why the budget is the next cue's onset rather than the cue's own window, and audio of three overlap policies. Aug 19, 2026 · 7 min
- 07 Kokoro's 24 English voices, measured: pitch, brightness, and pace Measured fundamental frequency, spectral centroid and speaking rate for every English voice in Kokoro-82M, a circulating description that does not match the audio, and where the model's speed parameter stops delivering. Aug 19, 2026 · 5 min
- 06 Kokoro-82M in the browser on an M1 Max: WebGPU, the WASM fallback, and what each costs Measured real-time factor, cold load time, and footprint for on-device TTS in Chrome, including a WASM fallback that runs slower than real time, a quantized model that is slower than full precision, and audio samples from every engine evaluated. Aug 19, 2026 · 11 min