-
The Coverage Lock: What Next-Token Prediction Can and Cannot Teach Multimodal LLMs About the Visual World
On why next-token prediction can make multimodal systems useful without fully aligning language with the visual world.
-
The New Locus of Representation? A Researcher's Perspective on the Large-Model Era with Limited Compute
A short take on where representation learning lives in the era of large multimodal models and limited compute.
-
The Shimmering Gap
Reflections on minds, intelligence, and knowing.
-
When the Internal Map Gets Warped: Why Identifiable Representations Are Crucial for Responsible AI
Why identifiable representations are crucial for responsible AI.
-
The Generalization-Specialization Dilemma
On the tension between flexibility and precision in biological and artificial systems.
