写作——托马斯·布雷迪|最后更新: 2026-5-27URLthbrdy.dev/writing/StatusLikeSavedMay 27, 2026 09:27 AM链接Writing From Transistor to Token Tracing the inference stack from Apple's Neural Engine through the M5 Max to autonomous experimentation — making each layer's real architecture visible. The Wrong Axis Mar 2026 An Anti-Debate on AI governance revealed that both participants were arguing on a map that doesn't match the territory. The real contest is between two industry coalitions fighting over whose values prevail in the rubble of the republic. Trust Topologies Mar 2026 Three biometric devices, three data flow paths, three trust models — all converging on the same protocol. The hard problems aren't data engineering. They're decisions about what counts as the same thing, what boundaries data should cross, and when the moment of measurement actually is. The Body Knows First Feb 2026 A biofeedback-assisted state navigation app that trains interoceptive awareness — helping you notice what your body already knows, in the middle of a life. The Circuitry of Science Feb 2026 Scholion extracts the logical architecture of research arguments — typed dependencies, implicit warrants, load-bearing claims — and makes them machine-readable. A methodology tested on clinical medicine, built for safety cases. The Valley of Death Is a Legibility Problem Feb 2026 The gap between research and product isn't a funding problem, a talent problem, or a culture problem. It's a structural failure: coupled questions answered sequentially, through frameworks that make the most important information disappear. What If You Could Feel Someone Breathing From Across the Room Feb 2026 On nervous system co-regulation, the science of physiological synchrony, and a research program for making interpersonal resonance tangible through consumer hardware. You're Already Feeling Something You Haven't Noticed Yet Feb 2026 On interoception, contemplative practice, biometric feedback, and building a tool that tries to become unnecessary. ML Compilers Are Leaving 15–30% of Training Throughput on the Table Feb 2026 Every major ML compiler faces the same constrained optimization problem. Every production compiler makes these decisions sequentially. What if a learned planner could make them jointly? Safety Cases Are Dependency Graphs That Nobody Maintains Feb 2026 Every organization building frontier AI systems faces the same structural problem: assembling interdependent claims into a coherent safety case. The dependency structure exists — but it lives in researchers' heads. Absolute Beginners++ Jan 2026 Everyone's teaching you about AI. Nobody's teaching you how to think. Five frameworks, five domains, eight decades — all describing the same structure.
