How to apply for ML Research Scientist

About Theorem

Theorem is an AI company that builds systems for formal program verification, aiming to improve software reliability. It operates as a programming languages and AI lab, creating products that scale correctness feedback for AI software engineering at arbitrary complexity. The team includes engineers and researchers with backgrounds in cryptography, infrastructure, security research, and top-tier companies.

About the role

As an ML research scientist, you will conduct research and develop products that enable AI models to perform software engineering tasks far beyond current human capabilities. You will work on problems several orders of magnitude larger, faster, or harder than state-of-the-art, and solve last-mile technical challenges. This role is part of a team building systems for formal program verification, directly impacting the reliability of all software.

A typical day

A typical day might involve researching new methods for scaling correctness feedback, implementing prototypes, and collaborating with engineers to integrate research into products. You could also spend time analyzing last-mile technical challenges and running experiments to push beyond state-of-the-art. The specific daily routine is not described, so ask about team structure and typical workflows during interviews.

Who Theorem is looking for

• Has a strong research background in machine learning, with a focus on program synthesis, formal verification, or automated reasoning.
• Experience with programming languages, type systems, or formal methods, and applying AI to software engineering tasks.
• Track record of working on large-scale, complex problems that push beyond state-of-the-art, with a focus on last-mile technical challenges.
• Comfortable collaborating with engineers and researchers from diverse backgrounds, including cryptography, infrastructure, and security research.

Tips for this application

• Highlight specific projects where you applied ML to program verification, synthesis, or software engineering tasks at scale.
• Demonstrate familiarity with formal methods or programming language theory in your resume or cover letter.
• Emphasize experience solving hard, last-mile technical challenges, not just high-level research.
• Mention any work on problems orders of magnitude larger or faster than state-of-the-art, with concrete metrics.
• If applying for the research fellowship, clarify your availability for a 3-6 month period and your specific research interests aligned with Theorem's focus.

What to cover in your cover letter

In your cover letter, focus on: (1) your experience applying ML to formal verification or program synthesis, (2) a specific example of solving a last-mile technical challenge, (3) your ability to work on problems orders of magnitude beyond state-of-the-art, and (4) why you want to join a lab focused on scaling correctness feedback for AI software engineering.

Research before applying

• Read Theorem's website and any published research to understand their approach to formal program verification and AI.
• Look up the backgrounds of the team members, especially those in cryptography, infrastructure, and security research, to understand the team's expertise.
• Investigate recent advances in formal verification and program synthesis to see how Theorem's work fits in.
• Find any talks or interviews by Theorem's founders or researchers to understand their vision and technical challenges.

Likely interview topics

Based on the job description, expect questions about:

• How would you apply ML to formal program verification at scale?
• Describe a time you solved a last-mile technical challenge in a research or engineering project.
• What are the limitations of current AI models for software engineering tasks, and how would you address them?
• How do you approach problems that are orders of magnitude larger or faster than state-of-the-art?
• Discuss a research paper or project related to program synthesis or formal verification that you admire and why.

Common mistakes to avoid

• Focusing only on general ML experience without connecting it to formal verification or program synthesis.
• Claiming to have solved problems orders of magnitude beyond state-of-the-art without concrete evidence or metrics.
• Ignoring the programming languages and AI lab aspect of the company, and treating it as a generic AI role.