EntroπaLabs
Writing

How each of these actually works.

One post per application in the playground: what it is for, the mechanism underneath, and the part that turned out to be harder than it looked. Written from the source, with the limitations left in.

How the Gaussian splatting viewer works

A radiance field is millions of fuzzy ellipsoids. Getting them on screen at 60fps is mostly a sorting problem, and the sort is the interesting part.

Perception · 3D

Reading a checkpoint in the browser

safetensors is a length, some JSON and a wall of bytes. A torch .pt is a ZIP with a pickle in it — which means writing a small pickle VM.

Language models · Interpretability

Deciding which experts to drop

Most of an MoE's experts are underused and several are near-duplicates. Turning that into a defensible pruning plan needs two numbers, not one.

Language models · Efficiency

A 3D renderer in 120 lines of canvas

No WebGL, no matrices library, no dependencies: project, cull, shade, sort. The robot playground draws every frame the way graphics worked before GPUs.

Robotics · Embodiment

A forecast that grades itself

Anyone can publish probabilities. The interesting question is whether they mean anything — so this one scores itself every night and refits.

Forecasting · Calibration

Valuing a hockey player from a box score

Points tell you who is good. Surplus value tells you who is *tradeable* — and the gap between those two is where a GM actually works.

Decision support · Valuation

The agent loop, and what it can't do in a browser

A tool-use agent is about forty lines of control flow. Everything difficult about it lives in the tools you are willing to hand it.

Language models · Agents