Writings ✍🏻

Why LLMs Are Non-Deterministic, and It Is Not Just the Temperature

Same prompt, different answer. The reasons turn out to be much older than LLMs — finite-precision arithmetic and parallel reductions that have haunted scientific computing for decades.

How Modern CPUs Talk — Part 1: The PCIe Highway to Internal Server Storage

The first time I saw a server chassis wide open, I was overwhelmed. It turns out the logic underneath is surprisingly clean — here is how modern CPUs talk to storage, and why PCIe lane budgeting shapes every server we build.

Built to Impress? — How Vibe Coding Fabricates Credibility

A 20,000-star GitHub repo promises WiFi-based 3D pose tracking through walls. Under the hood, it is an AI-generated shell with no real signal processing — and a near-perfect case study of how vibe coding exploits our engineering trust heuristics.

You're Probably Too Late to Build an AI Agent Framework — Here's Why That's Great News

The AI agent architecture race is over. The dust has settled on ReAct loops, context engineering, and Skills. If you missed the chaotic sprint of 2024-2025, you arrived at the perfect time.