Posts Tagged "AI"

A Knowledge Substrate With No Opinions

A follow-up to the harness I said I'd have to build myself: not the harness yet, just the floor it stands on. Every knowledge system drifts toward judging what it holds, a little inference here, a little reconciliation there, until the storage layer becomes a judge without anyone deciding it should. This is the discipline against that: a medium of record that certifies form and never truth, built to let a contradiction sit unresolved for as long as it needs to.

A Harness Has a Tribe

LLM Wiki spread fast because it names an old pain: everyone tries to organize what they read, and everyone quits. Strip the hype and it's a file format and three verbs, honest about being nothing more. Everything that decides whether it works is the harness it doesn't specify, and a harness is never neutral. The ones people are building encode what engineers know. That's why they broke on serious work in the humanities. Serious intellectual work needs its own harness, built from its own epistemology.

The Idols in the Machine

For fifty years, the valuable skill was speaking the machine's language. Then the machine learned to speak ours, and the scarce skill quietly became the oldest one of all: reading a fluent, confident, faintly sycophantic oracle closely enough not to be fooled by it. We have been here before. The medieval scholar lived where we live now, surrounded by authoritative texts he could not take at face value, and built an entire technology to interpret them. Then his whole project collapsed into, not for lack of intelligence, but for a hidden flaw in how he decided what to believe. That is the warning. The task now is not to kneel to the new oracle but to find the idols it plants in us and break them, so that we do not repeat the mistake that made the last people who tried obsolete.

From Authors to Curators

Hollow code. Unreadable documents that communicate nothing. The volume is growing and the quality is falling. The tactical fixes — better prompts, better models — aren't working, because this isn't just an engineering problem. The thinking happened in the struggle to produce an artifact. We removed that friction with AI, and took the thought process with it. We became curators rather than authors without noticing. To truly own what we produce, we have to take authorship back.

پیش از آنکه دیر شود...

نامه‌ای به همکاران و دوستان قدیمی درباره آینده امنیت سامانه‌های قدیمی در عصر هوش مصنوعی. دعوتی به بازاندیشی در دفاع نامتقارن، پیش از آنکه قواعد بازی برای همیشه تغییر کند.