Output
Engineering
MemberoftheTechnicalStaff,Interpretability
Neural analysis suggests this role is
optimal for Mid+ candidates.
“Member of the Technical Staff, Interpretability at Output. Skills: Model interpretability, Representation analysis, Probing methods. Develop methods for probing model representations. Understand how model encodes biological information”
Industry & Context.
Reverse-engineering; Experiment design; Data analysis
What They're Looking For.
Must Have
2+ years post-doctoral or industry research experience, 5+ years hands-on research and engineering experience, Proficient in Python, Proficient in PyTorch, Experience with large models on GPU infrastructure, Production-quality code, Well-tested and maintainable code, Comfortable working in shared codebases, Version control, Code review
Nice to Have
Background in chemistry, Background in biology, Background in computational biology, Background in biophysics, Experience interpreting ML models, Experience building visualization tools, Experience building analysis tools, Experience with multimodal models, Experience with multimodal representations, Contributed to open-source ML projects
What You'll Do.
Develop methods for probing model representations
Understand how model encodes biological information
Design experiments to identify capabilities
Map model learning about molecular interactions
Map model learning about biological function
Build methods to extract biological understanding
Create tools connecting model internals to concepts
Make model reasoning interpretable
Feed interpretability findings back into model development
Own pipeline from probing experiments to tools
Build robust systems on distributed infrastructure
How You'll Work.
Team & Collaboration
Work closely with pretraining teams; Work closely with generation teams
Full Job Description
Output has built a biological reasoning model that understands biology at the scale and complexity life actually operates. Our model independently learned the principles of molecular interactions, opening up drug treatments that were previously impossible. We're already generating therapies that traditional approaches cannot reach. The hardest problems in both AI and biology are being solved here, and there is room for you to own one. Output is currently in stealth, operated by a team of repeat founders and biotech veterans with multiple exits in AI x Bio, and backed by top-tier VCs including Y Combinator. You will continue developing methods to understand what our foundation model learns about biology, and build the tools that make it a glass box model. We believe that in biology, a model's reasoning must be visible. And the features you find are not just explanations: they expand what the model can do. - You will continue developing our methods for probing and reverse-engineering the model's learned representations, understanding how it encodes biological information across molecular scales - You will design and run experiments to identify and characterize capabilities, mapping what the model has learned about molecular interactions and biological function - You will build methods to extract the model's biological understanding as explicit, usable outputs that downstream systems and researchers can act on - You will create tools that connect model internals to meaningful biological concepts, making the model's reasoning interpretable to scientists and useful in practice - You will work closely with the pretraining and generation teams, feeding interpretability findings back into model development to strengthen the capabilities you uncover - You will own the full pipeline from probing experiments to production-quality interpretability tools, building robust systems on distributed infrastructure About You - You have a PhD in computer science, machine learning, physi
Applying for this Member of the Technical Staff, Interpretability role?
Most applicants get filtered before a human reads their resume. See if yours makes the cut.
How to Apply on Ashby
- Ashby is a fast modern ATS — most applications take under 3 minutes.
- The resume parser is strong; verify parsed experience dates and job titles.
- Custom screening questions are often scored algorithmically — answer completely.
- Location field affects geo-based screening; use your actual metro area.
ANONYMOUS · UNFILTERED
What do employees actually say about Output?
Real rants from real employees. Read before you apply.