Livgens

How anything real is represented.

Notes on representing a thing exactly, in sound, light, matter and motion, and on the field around it: where real data runs out, and what it takes to fill it with something that holds up.

Livgens · blog RSS
01Latest
AllDataRepresentation3DMaterialsSoundHealthMeasurementSimulationAgents
Field notes · 21 October 2026Wearables and the truth: what the 2025 validation studies sayIndependent studies in 2025 measured consumer rings and watches against ECG, polysomnography and laboratory VO2max. Some readings are close to the reference and some are far from it, which is why every number a wearable shows should carry its own grade.
Research · 21 October 2026The data you need was never collectedPublic text is running out, robot data is scarce, and models fed their own output lose the rare cases first. The way through is to represent real things exactly, run them forward, and hold every result to the world.Field notes · 21 October 2026Synthetic data in 2026: what it is good for, and where it breaksSynthetic data became ordinary infrastructure in 2025, from open post-training sets to world models for physical AI. The research on model collapse says where it breaks, and points to generating from a measured representation of the thing rather than from a model of the data.Field notes · 21 October 2026Restoring soundTaking noise out of a recording, balancing it, matching its tone and setting its loudness are acts of representation: each one claims to know what the sound really was. Looper makes those claims inside Ableton Live, and shows what it changed.Field notes · 21 October 2026Representing a garmentA garment is a body, a pattern, a cloth and the way they settle together. Representing it exactly means carrying every one of those as a measurement, and saying plainly which parts have been checked against real cloth.Field notes · 21 October 2026Measuring a bodyWearables report some things well and others badly, and every reading carries noise. The question that matters is whether a change you made is working for you, and that can only be answered on your own measurements, at your own measurement error.Field notes · 21 October 2026Fashion goes three-dimensional, and the law wants the facts3D design and digital sampling spent a decade making garments look right on a screen. European rules on textiles now ask each garment to carry facts about what it is made of and how it ends its life, which raises the value of a 3D garment that is a measured record rather than a picture.Field notes · 21 October 2026Capturing materials: measure first, then generateIn 2025 and 2026 graphics settled on shared ways to describe materials and captured scenes, and generative models learned to paint physically based materials onto any mesh. The formats and the generators are ready; what decides whether a material is true is still a measurement.Field notes · 21 October 2026Bodies in three dimensions: from plausible to measuredOpen body models became far broader in 2025, covering every age and separating skeleton from shape, and recovering a body from one photograph is now routine research. Fit, sizing and health are decided in centimetres, which makes the measured body, with its error stated, the representation that matters.Field notes · 21 October 2026Audio AI in 2026: licensed, labelled, and measuredGenerated music now makes up more than half of the new tracks arriving at one major streaming service, and the labels have moved from suing the generators to licensing them. The harder problems left are provenance and measurement, and both come down to knowing exactly what a sound is.Research · 21 October 2026A little real data goes a long wayLearn from a checked simulation first, then from a few real measurements, and each measurement counts for more. A registered Living Worlds study on unmetered power feeders shows how much more, where the advantage ends, and what happens when the simulation is wrong.Field notes · 21 October 2026A coding agent in the browserCoding agents now write a great deal of code. The scarce thing is evidence that the code works. Livgens Programmer reads first, shows every change, runs the tests and reports what passed and what did not, in the terminal and in a browser workspace that keeps your work between visits.