Visualizing model's thought
2025-7-10
With the advent of dimension reduction algorithm, one can visualize model's thought, especially on those words that have geometry semantics.
Blog
Blogs by myself.
Randomness in language model
2025-9-10
This blog explores the concept of randomness in language models, discussing its implications and potential applications.
VisionGuard: vibe-coding a dry-eye relief app
2026-5-9
A native macOS app I built with Claude Code over a weekend — periodic rest reminders, blink detection (with external camera support), and f.lux-style color temperature, bundled together.
How a 1-hidden-layer ReLU network learns a continuous function
2026-5-9
A geometric construction for the universal approximation theorem in 1D: pairs of ReLU neurons make ramps, and stitching them yields a piecewise-linear interpolant. An empirical check on $f(x) = x^2$ confirms a trained network does exactly this.
