TLDR;
We’re hiring Research Scientists to work on Ambitious Mechanistic Interpretability (AMI) at Principles of Intelligence: developing tractable and realistic data structure models, quantifying how the features that AI systems learn are rooted in data, and developing synthetic datasets to benchmark and improve interpretability tools. Fully remote. Strong motivation to work in AI Safety required. Apply here by 21 June.
Position Details
Location: Fully remote. Candidates with availability during ET core hours are strongly preferred.
Visa / Work authorization: We can provide visa sponsorship for candidates based in (or relocating to) London (UK) and Berkeley / Bay Area (US), depending on candidate circumstances and role fit.
Type: Full-time.
Start date: July 2026.
Salary range: Salary is dependent upon location and experience. Our estimated range for research scientists is $100,000-$250,000 gross.
Deadline to apply: Please submit the application form by 21 June. Applications will be reviewed on a rolling basis.
Referrals: We’re happy to receive referrals. Please submit recommended candidates via this form.
Core Responsibilities
We are hiring for Research Scientists to work on Ambitious Mechanistic Interpretability at PrincInt. The work is aimed at building and advancing methods to:
- Develop tractable, realistic, scale-aware data structure models.
- Quantify the relationship between data structure & learned features.
- Develop synthetic datasets to benchmark & improve interpretability tools.
The team is currently focused on exploring a data model based on high-dimensional percolation theory which encodes a statistically self-similar and sparse structure.
This role is fully remote, so we’re looking for people who can thrive in an async environment and are genuinely excited to work on AI safety.
Requirements
We’re looking for a research scientist to lead research projects at the intersection of physics and mechanistic interpretability. You’ll work with a small team conducting interdisciplinary research on synthetic data models, bridging the gap between theories of learning and practical AI interpretability tools. Strong candidates are motivated to contribute to AI safety and can collaborate effectively in a fully remote environment.
Research Scientist
- Must have
- PhD or equivalent research experience in a technical field such as math, physics, neuroscience, computer science, or complexity science
- Ability to communicate complex, novel ideas to technical and non-technical audiences
- Working knowledge of neural networks and deep learning
- Experience with scientific computing in Python
- Comfort working in a fully remote environment
- Strong motivation to contribute to AI safety
- Great to have
- A history of publications in ML or another quantitative field demonstrating a track record of original, impactful contributions
- Working knowledge of statistical physics, especially percolation theory and renormalization; mechanistic interpretability; stochastic processes, especially coalescent processes; combinatorics; or network science
- Hands-on experience with fractal geometry; neural scaling laws (theoretical or empirical); graph and tree algorithms; or representations and theories of concepts in cognitive science
- Familiarity with AI safety research engineering, i.e. content in the ARENA curriculum, or enough ML experience to quickly pick it up
We don’t expect anyone to cover every field above. If this research excites you, apply.
Benefits and Salary
Your starting salary range depends on role, prior experience, and location. Our estimated range is $100,000-$250,000 gross.
Our benefits include:
- Time off: 4 weeks of paid holiday per year, plus additional personal time off.
- Flexible working hours: Flexible working hours with an emphasis on outcomes and reliable coordination with teammates.
- High ownership & autonomy: A high-trust environment where you’ll own your work end-to-end, with meaningful responsibility and room to improve systems.
- Employment type: We’re open to either full employment or a contractor arrangement, depending on location and candidate preference.
- Team meetups: We get together in person 2-4 times a year for focused work time, great conversations, and the kind of momentum you only get when the whole team is in the same room.
Application Process
- Application: Submit an application by 21 June, 2026.
- Screening call: Attend a brief screening call, where you’ll have the chance to ask questions about the role.
- Interview: Attend a remote interview to assess technical knowledge and team fit.
- Research Talk: Give a presentation to the team on your past work.
- Work trial: Attend a paid one-day remote work trial where you’ll complete a research task and meet the team.
- References: Share references who can comment on your aptitudes. Good references can notably strengthen your application.
Diversity and Inclusion
We’re aware that factors like gender, race, and socioeconomic background can affect people’s willingness to apply for roles for which they meet some but not all the suggested attributes. We’d especially like to encourage people from underrepresented backgrounds to express interest.
There’s no such thing as a “perfect” candidate. If you’re on the fence about applying because you’re unsure whether you’re qualified, we’d encourage you to apply.
If you require any adjustments to the application process, such as accessibility accommodations, additional preparation time, or other, please contact [email protected]. We’re happy to support your needs and adjust the application process.
