dmodel

look inside the model

About

d_model is a fundamental AI research lab. We're focused on automating interpretability and, more broadly, the science of deep learning — building the tools that let humans understand and steer the systems they train.

Other vendors build hackable environments for specific tools. We work on models themselves — making them more understandable[1][2], more cooperative[3], and more aligned.

We think alignment is a real and unsolved problem, but an empirical one. The path forward looks like serious experimental work, layers of inoculation, and automating the research itself[4] — a marriage of science and philosophy.

If you want to work on the parts of the problem that matter the most, talk to us.

April 2025

Inside the CodeBot

A Gentle Introduction to How LLMs Understand Nullability
March 2025

How Language Models Understand Nullability

We study how models represent the nullability of program values. We measure how well models of various sizes, at various training checkpoints, complete programs that use nullable values, and then extract an internal representation of nullability.
September 2024

Steering Characters with Interpretability

We think you can make better characters with steering vectors. Try it out in our notebook, or check out some of the examples from the screenshots in the post below.

Team

anish co-founder
dmoon co-founder
alex research
adam research
autumn research
curry chief of staff
jeff research
bishka research
ritesh research
tyra research
sharuya research

If you're interested in joining our team, please reach out.