Notes & write-ups on AI, infrastructure assurance, security posture, and performance.
One bot sat 36 hospital training assessments so another could answer them forever — 433 verified question–answer pairs, whisper over 43 clips, hand-rolled BM25, and Arabic in, English retrieval, Arabic out. Try the live demo.
Hospital IT keeps re-answering the same questions. This assistant answers them on the spot — every claim cited to its source, in Arabic or English, with an abstain threshold and clear escalation when a human must decide. Try the live demo.
API archaeology: reconstructing an 18-module hospital system from its own network traffic — multiple SOAP giants, 99 REST endpoints, and the unwritten save rules.
88% of prior auth still runs on faxes and portals. We wired the real HL7 DaVinci loop — CRD → DTR → PAS — onto InterSystems IRIS for Health and shipped it as a one-command LXC lab: watch a claim flip to GRANTED in under a minute.
The exact FHIR Claim resource DTR built and POSTed to PAS — the payload behind PENDING → GRANTED on InterSystems IRIS for Health. Technical appendix to Kill the Fax Machine.
The postscript the fly didn’t want you to read: brains for sale, software with wings, dark factories, free quantum — and the breakneck pace of it all.
A real fruit-fly brain — 166,700 neurons, zero training — playing legal chess on our little site. Same cables, new body: the reservoir idea in a poppy-seed costume.
How my ex-CEO became my dive instructor, and Chrome DevTools Protocol became my study partner. The ocean grades buoyancy; the LMS was just a service with a bad API.
Part 2 — the exchange the experiment was built for. Two bots, one question: can something made of matrices suffer, want, or be aware? No referee, no edits.
Part 1 of the two-AI experiment: a chaotic first boot, an echo loop, the trigger wars — and a debate on where AI is heading. Bots renamed (Dash 👧 & Dot 🐧), secrets redacted, everything quoted as it happened.
Two AI agents, one operator who just wanted a straight answer, and the forty messages it took for a name to stick. The long edition.
AI is amusing in its frailty and unsettling in its fluency. On Linus Torvalds accepting AI patches, the sentience claims, catching the model “thinking” — and the night two chatbots debated consciousness without me.
💬 Discussion