About

Latent Lexicon is a catalog of newly named abstractions: hidden structures beneath seemingly unrelated phenomena, surfaced by AI models and reviewed by people.

What we do

Large language models have been trained on a vast share of the text humanity has digitized, spanning domains, cultures, and centuries in ways no single human mind could. We suspect that in doing so they have formed latent abstractions: deep structures beneath seemingly unrelated phenomena, in how the world works and in how we experience being alive. Some may capture what humanity has long sensed but never named. Others may be genuinely new, and those are the ones we care about most.

We work with AI models to propose candidates, review them by hand, give each a name and a precise definition, and add them to a shared vocabulary. Each entry in the Catalog names one such abstraction and shows what it explains, across fields that rarely speak to each other. Our Methodology explains how we find and review them, and our Mission explains why we believe this matters.

How it started

Latent Lexicon began with a question: what is fundamentally happening inside large language models? In January 2026, that question became an essay, The Mathematical Mirror: LLMs and the Great Unknown. The essay led to two talks: one at z.ai's headquarters, at the invitation of its co-founder and chief scientist, Professor Jie Tang, and one at Tsinghua University, at the invitation of Professor Juanzi Li, Xinru's former advisor.

Preparing those talks made the idea behind this project clear: our ability to recognize and name patterns is a large part of what gave us technology and civilization, yet profound patterns still remain unnamed. By August 2026, Latent Lexicon had become a full-time project.

Who we are

Latent Lexicon is built by two former Google engineers.

Xinru Zhang: Co-founder. Spent six years as a software engineer at Google. Studied computer science at Tsinghua University, researching knowledge graphs and entity linking, the study of how words connect to the things they mean. Now works on Latent Lexicon full-time, writing about AI, language, and how minds turn experience into concepts. LinkedIn · GitHub

Co-founder: A former Google engineer whose talent we deeply admire, who prefers to remain anonymous and let the work speak for itself.

Get in touch

We're building Latent Lexicon in the open and welcome ideas, critiques, and collaborators. Write to us at contact@latentlexicon.org.