Engineering notes, AI tools, and the occasional thought from the workbench.
Procedural Graphs store steps, conditions, and common failure modes in a workflow that can guide long-horizon agents locally and improve through validation.
read note →How GPT-Live-1 handles speech, pauses, and interruptions in one model while delegating complex work to a backend model.
read note →Bypass Gmail's front-end UI limits and batch-mark thousands of unread emails via Apps Script, with rate-limit-safe pagination and progress logging.
read note →A small technical station for keeping the useful parts of ongoing work in public.
read note →Twelve communication tools—from reframing premises to calibrating uncertainty—visible in Dario Amodei’s answers to hostile questions.
read note →Technology needs high-level boundaries: a victory achieved by destroying the values being defended is not a meaningful victory.
read note →As information processing gets cheaper, work grounded in the physical world and human trust becomes more valuable.
read note →Three-second content culture turns complex arguments into reflexes and leaves less room for careful reading and mature discussion.
read note →Disagreement can be repaired; collaboration ends when honesty, motives, and shared intent are no longer credible.
read note →How ElevenLabs grew from a text-to-speech wedge into a production stack for dubbing, localization, and real-time voice agents.
read note →From latent diffusion to multimodal visual intelligence, with deployment feedback shaping the Flux model roadmap.
read note →Why Luma AI moved from 3D capture to video and then toward unified models for end-to-end multimodal work.
read note →AI scaling, context feedback loops, infrastructure bottlenecks, and why compute remains in a pre-standardization era.
read note →