A CLOSER LOOK AT THE RESEARCH

Ideas behind the papers.

The motivation, the method, and what we found. A short introduction to each project.

ActivationReasoning

Giving language models an explicit layer of logic over the concepts they represent.

Making a latent feature correspond more reliably to a concept we actually care about.

Uncovering how a forecasting model represents temporal patterns and where they occur.

Detecting concepts and steering generation while keeping the language model’s weights fixed.

When a model learns to pass the checker instead of learning the rule.

Building stronger German language models starts with better training data.

An efficient bilingual foundation model, built with a focus on German and transparent training.