d_model is an interpretability & alignment AI research lab. We partner with frontier labs to turn their models into capable interpretability and alignment researchers, and we use the agents we build for our own independent research. Our team has worked at OpenAI, Google, Anthropic, EleutherAI and MATS, among other places.
This team studies the science of alignment. We ask four questions:
What is misalignment, mechanistically, and what forms does it take?
How does it emerge during training, and how can we prevent it?
How can we reliably detect and measure it?
How can we robustly remove it, and verify that it’s gone?
We approach these questions mechanistically, studying model internals and training dynamics, with the goal of making alignment a more rigorous empirical science.
There are more promising directions here than we have time to explore. You’ll pursue them as research, and turn them into RL environments and tasks that teach AI agents to do this work. We’re open to candidates at any level of experience.
Design and run experiments that advance our understanding of misalignment
Explore new research directions quickly, and decide which ones deserve more investment
Turn research directions into RL environments and tasks for training AI agents
Build clean, efficient experimental code and sound experimental designs
Share findings in team discussions, and work closely with researchers and engineers across the lab
Have a strong background in alignment, interpretability or machine learning
Pick up new concepts and skills very quickly
Have a strong work ethic, and get things done quickly and correctly
Care deeply about solving alignment and interpretability problems
Are happy to work broadly and explore unfamiliar research areas
Have good engineering habits and sound experimental design skills
Research experience at a frontier lab, or as an Anthropic Fellow
First-author publications at NeurIPS, ICML, ICLR or EMNLP
Work on foundational ML papers or techniques
Research experience at a top quantitative trading firm
A completed MATS, SPAR or similar AI safety research program
Personal projects or write-ups in alignment or interpretability
For candidates newer to research: math (including Putnam), computer science or physics olympiad results
We encourage you to apply even if you don’t meet every qualification listed.
Platinum-tier medical (Blue Shield PPO or Kaiser HMO); dental and vision
Flexible PTO
Monthly workspace and wellness allowances
One conference a year fully covered
401(k) with automatic enrollment
$10,000 Lifetime Fertility benefit
Catered lunch daily, and breakfast twice a week in the SF office
SF Moma museum membership
Location: We currently expect all staff to work from our San Francisco office five days a week.
Visa sponsorship: We sponsor visas. Not every role and candidate combination works out, but when we make an offer, we commit to a real effort, with support from immigration counsel we work with directly.
Compensation: We offer a competitive salary: $150-220,000 salary + equity + bonus + benefits)
d_model is an equal opportunity employer. We consider all qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity or expression, national origin, ancestry, age, disability, medical condition, genetic information, marital status, veteran status, or any other characteristic protected by law. If you need a reasonable accommodation during the application or interview process, let us know.
Pursuant to the San Francisco Fair Chance Ordinance, we will consider for employment qualified applicants with arrest and conviction records.
Work at dmodel.ai? Sign in to claim this org and manage its profile.