measured
14 stories and discussions about measured, aggregated from every source we track.
Waves, wind, water and tides from NOAA's network of buoys, coastal stations, tide gauges and estuaries, from the Bering Sea to Samoa, charted and analysed.
I review more code than I write now. I'm not sure I'm getting better at it. It's a small,...
Gemma 4 E2B q4_0 served by llama.cpp on one laptop, CPU-only and on a GTX 1650 Ti, rebuilt on CUDA 13.4 and re-measured in CPU, GPU, GPU, CPU order with a temperature gate. The card takes decode by 4.14x, and run order moves the answer by about 2%.
Last week we wrote about the classification problem hiding in your LLM bill, and about how to read a...
During the Apollo missions to the Moon, a critical task for NASA was determining the spacecraft's position. To accomplish this, they devel...
A few weeks back I watched an agentic pipeline burn a month of budget in three days. The instinct...
A real app on the cheapest droplet money buys: 1,117,976 requests with zero errors, 9,737/s static and 18,557/s once nginx cached the dynamic route. The limits I hit were a config default and my own arithmetic, not the hardware.
What happens when enterprise requirements - human approval gates, audit trails, structured output -...
I've run tmux since somewhere in 2016 because I can script it. Ten years of scripts later my status bar was using 15.4% of a core on battery, so I measured it, cut it to one call, and moved the scripts into one Rust…
deepseek-v4-flash from DeepSeek's API vs through OpenRouter (Cloudflare, Together and DigitalOcean), measured daily from 6 cities: direct 698 ms, routes 721 ms to 774 ms.
Étude originale : 91,8 % des 400 sites .fr mesurés ont un SPF, 69,2 % un DMARC, mais 28,8 % seulement l'appliquent. Méthode, marges d'erreur et limites.
I benchmarked the Jev model against an LLM agent on real browser navigation: 13x faster per task, and one setup mistake that broke every single test.
I tracked the endings of 7,384 films, and measured the emotions on 1.57 million actors’ faces, to find out whether the movies still send us home happy.
We ran three context-compaction strategies through the same 178-question exam. FutureOS kept 83% of the values that had been compacted away; OpenCode kept 47%, Codex 38%. The difference isn't a better summary — it's…