Live speech into captions that show who is speaking, when each word is said, and how it was said. A headset with a mic array adds where the voice came from, as a pulse beside the ear.
Jin-Myung-Lee
Users can no longer build a mental model by recombining finite rules, so the agentic OS needs criteria that CLI and GUI theory cannot supply.
Why originality and persuasiveness pull against each other in natural-language argument. Universal evidence exists in formal proofs; in prose, rarer claims have rarer support.
On editing a senior-to-junior essay tradition from 68 to 188 pages. What trusting collaborators looks like under year-end deadline pressure.
On why a graduation speech only matters if it could only come from you. If TTS could read it without losing anything, the speaker wasn't needed.
A new metric for evaluating machine unlearning in generative adversarial networks. Existing scores miss how forgotten classes influence their neighbors.
Thirty-three handwritten syllables in, a Light, Regular and Bold Korean font family out. What began as a Django class project now ships as a desktop app for macOS and Windows with the model inside.
Effective resistance in graphs, via the Laplacian matrix and its pseudoinverse. Worked examples from a 2-edge circuit to a Python solver on the cube graph.
Why we struggle to see AI as a material system of hardware and mathematics. Media narrative and fluent language convince us a black box has a mind.
A cathode ray glows green not because electrons are green, but because of the gas they hit.