Hey there

Welcome to my blog

Why does Muon work?

I wanted to understand why the Muon optimizer works, so I went back through SGD, momentum, Adam and Shampoo and built a simulation for each idea.

November 10, 2026 · 25 min · npow

The State of Robot Intervention (Nov 2026)

Where the money is going, what teams are hiring to build, and which companies could own the operating relationship.

November 10, 2026 · 11 min · npow

How Well Calibrated Are Jev’s Probabilities?

Jev is an AI forecasting service that assigns probabilities to yes-or-no questions. Each question can include true and false text defining what counts as yes or no. With no deadline and both fields set to empty strings, Jev gave 35% to bitcoin will reach 100 million and 14% to bitcoin will reach 1 million, averaging 100 calls each. Calibration asks whether forecasts assigned 20% come true about one time in five across many independent events. ...

September 21, 2026 · 6 min · npow

The State of Robotics (September 2026)

A September 2026 census of 2,538 robotics and physical-AI companies across 24 use cases, with funding, hiring, geography, deployment, and an interactive company directory.

May 31, 2026 · 2 min · npow

Does Caveman Mode Actually Work?

Caveman is a Claude plugin that asks the model to answer in terse fragments. Does it save money? Sometimes—mostly when the model would otherwise produce a very long answer. I compared Caveman’s lite, full, ultra, and wenyan modes with two baselines: no added instruction and “Answer concisely.” The benchmark covered 15 prompts, three runs per condition, and two Claude Opus models. I measured token cost, not answer quality. Model and task Caveman cost vs. concise Claude Opus 4.7, short Lite +10%; Full +3%; Ultra +3%; Wenyan +10% Claude Opus 4.7, long Lite +9%; Full +10%; Ultra −5%; Wenyan −1% Claude Opus 4.6, short All modes: 3–16% less; uncertain Claude Opus 4.6, long Lite −54%; Full −56%; Ultra −59%; Wenyan −58% Positive numbers mean higher cost; negative numbers mean savings. The practical rule: Caveman has to save enough output tokens to pay for its added instructions. That is easy with a long tutorial and hard with a short answer. ...

April 20, 2026 · 4 min · npow