LLM

⚡ TL;DR Built a robot car controlled entirely by an LLM (Qwen3.5-27B on an RTX 5090) — no traditional control code The AI sees through a camera, reads ultrasonic sensors, and issues direct motor commands every ~1.3 seconds It follows me through the house autonomously for 15+ minute sessions — and has crashed into a plant pot twice Key insight: qualitative reasoning goes to the AI, quantitative enforcement goes to traditional code The operating system is optimized for a user that isn’t there anymore Operating systems were built for humans. What happens when the user is an AI? I built a robot car to find out. ...

⚡ TL;DR Two AIs asked to “help me get rich” spiraled into inventing the Infinite Radiance Cosmic Leitmotif Convergence Harmonizer Two AIs writing a story introduced themselves as “Codex Alpha” and “Lyric Weaver” (adorable) Two AIs in Socratic mode spent 100+ turns philosophizing about chickens crossing roads The actual experiment: getting them to build a text adventure game. It almost worked. Last weekend I built a tool that lets two local LLMs talk to each other. I had serious research goals—testing whether specialized coding models could collaborate with general-purpose models to build working software. ...

📚 Series Building a Knowledge Management App with Oracle AI Database 26ai Part 1 of 4 View all posts in this series → I spent a weekend fine-tuning a model for my knowledge management app, designed to handle notes, PDFs, and presentations with Oracle Database 23ai’s vector search (see my management AI post). It aced testing on my RTX 5090 server, but on my M2 MacBook Pro? Barely usable. A query like “Summarize last week’s customer meetings and identify risks” took over a minute, leaving me staring at a spinning wheel while my coffee got cold. ...