Phuc Vinh Truong Articles https://www.phuc.net/articles/index.html Latest essays by Phuc Vinh Truong on AI, science, and Information Force Theory. en-us Mon, 23 Feb 2026 20:23:08 GMT Thought Experiment: If “God Is Real,” Can We Convince AI — and What Happens Next? https://www.phuc.net/articles/what-if-ai-believed-in-god/ https://www.phuc.net/articles/what-if-ai-believed-in-god/ Mon, 23 Feb 2026 00:00:00 GMT A systems thought experiment on defining God as the universe's orchestration layer, and how that frame changes AI reasoning, ethics, and governance. SWE-bench Isn't a Model Leaderboard. It's a Discipline Leaderboard. https://www.phuc.net/articles/swe-discipline/ https://www.phuc.net/articles/swe-discipline/ Mon, 23 Feb 2026 00:00:00 GMT People argue about which model is "best" at coding. That argument is already outdated. SWE-bench Verified rewards a system that turns a failing test into a passing test — with receipts. Pied Piper Was Fiction. Here's Runnable Compression With Receipts (PZIP). https://www.phuc.net/articles/pied-piper/ https://www.phuc.net/articles/pied-piper/ Mon, 23 Feb 2026 00:00:00 GMT Real compression isn't wizardry — it's a fixed corpus, a fixed command, a fixed stdout, and a decode that always round-trips. PZIP challenges you to stop selling trust and ship proof. Your AI Agent Has No Permission Slip — So We Built One https://www.phuc.net/articles/oauth3-standard/ https://www.phuc.net/articles/oauth3-standard/ Mon, 23 Feb 2026 00:00:00 GMT OAuth3 is the open standard for AI agent delegation — scoped tokens, step-up consent, evidence bundles, and synchronous revocation. If Your Math Agent Can't Check Itself, It's a Poet https://www.phuc.net/articles/math-dragon/ https://www.phuc.net/articles/math-dragon/ Mon, 23 Feb 2026 00:00:00 GMT Most math "reasoning" demos are storytelling — convincing, sophisticated, and wrong when you actually check. The Stillwater rule is simple: reasoning must be checkable, with witnesses, falsifiers, and receipts, not just plausible. Most Autonomous Agents Are Just Security Incidents Waiting for Wi-Fi. https://www.phuc.net/articles/malware-with-a-diary/ https://www.phuc.net/articles/malware-with-a-diary/ Mon, 23 Feb 2026 00:00:00 GMT If your agent can browse the internet and execute tools — and you treat webpage text like instructions — you didn't build an assistant. You built malware with a diary. The SECRET to Human Evolution DARWIN MISSED https://www.phuc.net/articles/how-humans-outsourced-their-minds/ https://www.phuc.net/articles/how-humans-outsourced-their-minds/ Mon, 23 Feb 2026 00:00:00 GMT An argument that humanity became dominant by externalizing memory, and that human-AI shared memory may form the next apex intelligence system. Happy Year of the Horse / AGI: A Message to Humanity https://www.phuc.net/articles/happy-year-of-the-horse-agi/ https://www.phuc.net/articles/happy-year-of-the-horse-agi/ Mon, 23 Feb 2026 00:00:00 GMT A personal letter on open-sourcing core AI work, choosing shared progress over hoarding, and inviting builders to create a more open, verifiable, and humane future. Game of Death: AI Tower Challenge https://www.phuc.net/articles/game-of-death/ https://www.phuc.net/articles/game-of-death/ Mon, 23 Feb 2026 00:00:00 GMT Inspired by Bruce Lee's tower, the Stillwater Game of Death maps 10 failure modes that break AI agents in production today across 5 floors of honesty, foundation, proving, precision, and mastery. Earn your belt by running the system, not just reading about it. The Most Important Skill Fathers Must Teach in the AI Age https://www.phuc.net/articles/fatherhood-in-the-ai-age/ https://www.phuc.net/articles/fatherhood-in-the-ai-age/ Mon, 23 Feb 2026 00:00:00 GMT I built an AI with my son and it changed what I think fatherhood means. In the age of AI, the scarce resource is not knowledge — it is discernment. AI Kung Fu for Humanity: The Stillwater Vows (Discipline Over Vibes) https://www.phuc.net/articles/ai-kungfu/ https://www.phuc.net/articles/ai-kungfu/ Mon, 23 Feb 2026 00:00:00 GMT In a world where AI acts instead of just answers, raw intelligence is no longer the advantage — discipline is. Stillwater OS is a dojo for AI work: verification over vibes, receipts over rhetoric, boundaries over bravado. A child can count better than Frontier AI Models https://www.phuc.net/articles/ai-cannot-count/ https://www.phuc.net/articles/ai-cannot-count/ Mon, 23 Feb 2026 00:00:00 GMT A 128K context window doesn't mean your model can count — it means it can read more while still doing approximate attention math. The fix is the Counter-Bypass Protocol: let the LLM classify, let the CPU aggregate. If Your Agent Obeys Webpages, You Don't Have an Agent. You Have a Remote-Controlled Exploit. https://www.phuc.net/articles/agent-is-exploit/ https://www.phuc.net/articles/agent-is-exploit/ Mon, 23 Feb 2026 00:00:00 GMT Most "prompt injection" advice is vibes: "don't get jailbroken", "be careful", "follow policy". None of that works. Here's the cure in four lines.