Reverse engineering intelligence with interpretability.

LAION

Building an interpretable VLM that maps text and vision towers on top of OpenCLIP.

KAIST

Using interpretability tooling for preference detection analysis in LLMs.

AEIA Lab

Training SAEs on a LLM to sense conflict, ambiguity, and tone, and choose the right response.

Zyogen

Building interpretable medical vision transformers and diffusion models.

Training a ViT on wound recovery imaging, then interpreting it to distill a prediction system.

Finding features that drive recovery signals, then building and distilling leaner predictor to be able to inspect.

Mapping correlations between the human brain and LLM neural nets.

Building a testbed for drugs and cures aimed at Alzheimer's, dementia, and schizophrenia.

Finding deceptive features for AI safety.

Locating and studying deceptive features before they become failure modes for AI safety.

Finding if models have mechanistic representation of their context exhaustion.

What happens inside when a model knows the window is ending, when it knows it is about to die.

Interpreting and quantizing world models.

Reading how a model represents the world, then compressing that structure so it stays inspectable and usable.