How machines represent meaning.
Semiotics is the study of signs and how they come to mean something. We take that question seriously as an engineering discipline: what a model has actually learned, how to surface it, and how to know when to trust it.
Representation and meaning
What structure lives inside learned representations, and how that structure can be read out as something a person can inspect, name, and challenge.
Grounded language systems
Methods for keeping generative models anchored to sources: retrieval, citation, and abstention, so a system can show where an answer came from.
Evaluation
Building evaluations that measure the thing you actually care about. Benchmarks age quickly; the discipline of writing good ones does the real work.
Notes
We publish working notes as they become useful to people outside the lab. If you want to be notified when notes go up, get in touch.