-
WorldTest: how do we know whether an AI has learned how a world works?
A linearized, story version of my TAIC'26 talk on benchmarking world-model learning with environment-level queries. With live environments you can play.
-
Grammars that generalize: trying to find domain-invariant bird recognition
Combining a small DSL with a neural network for image classification gives domain invariance for free.
-
Bayesian synthesis of probabilistic programs for automatic data modeling
Building a chain of mutated programs for synthsizing probabilistic programs