Writing
-
Technology Advances, Distribution Lags: The Structural Contradiction of the AI Era
AI is approaching zero marginal cost for cognitive labor, decoupling economic growth from employment. This is not the same story as previous industrial revolutions.
-
From Custom Agents to Agent Runtime + Skill: Convergence in AI System Architecture
A systems-level view of how AI applications are evolving from bespoke agent implementations toward a layered architecture of interchangeable Runtime and reusable Skills.
-
Intuition First, Explanation After: What Humans and LLMs Have in Common
A closer look at the shared tendency of humans and large language models to arrive at conclusions before generating plausible justifications — and what this means for how we understand "understanding."
-
Why Rust Makes Lifetime Annotations Part of the API
Rust lifetime annotations are not just compiler hints. They are part of the function contract, which is why explicit lifetimes matter for callers, traits, and API stability.
-
Why Temperature-0 LLM Inference Still Isn’t Deterministic
A practical look at why identical prompts can diverge under temperature=0, how batch-variant kernels create that behavior, and why batch-invariant operators matter for RL and reproducible experiments.