Position Description

TLDR: Timaeus is hiring a Research Scientist (Theorist) to extend the mathematical foundations of singular learning theory (SLT) and its applications to AI alignment. In-depth familiarity with SLT is not required.

We're looking for mathematical maturity (PhD or equivalent) with backgrounds like algebraic geometry, Bayesian statistics, statistical physics, information theory, or learning theory. You'd join our theory team working on directions like extensions of SLT to reinforcement learning, connections with Solomonoff induction, and the mathematical underpinnings of our spectroscopy and patterning research programs. Compensation $80k–$200k depending on location and seniority; remote-first with hubs in Berkeley, Melbourne, and London. Details on this and other open roles below.

1About Us

Timaeus’ mission is to empower humanity by making breakthrough scientific progress on alignment. Our research focuses on applications of singular learning theory (SLT) to interpretability and alignment.

Our two primary research programs are:

• Spectroscopy: Spectroscopy is an interpretability methodology that uses susceptibility-based methods to discover and characterize internal structure in neural networks, grounded in weight space (in the tradition of statistical physics and Bayesian statistics) rather than in activations (like sparse autoencoders and probing techniques). See Gordon et al. (2026).

• Patterning: Patterning is a methodology for steering what structures neural networks develop by inverting the spectroscopy signal to reweight training data. See Wang & Murfet (2026).

In addition to this, we conduct basic research on foundations in deep learning theory: on extensions of singular learning theory to reinforcement learning, on connections with Solomonoff induction and related areas, on improving and scaling the MCMC methods that underpin spectroscopy and patterning, and on further developing the connections between data structure, loss landscape geometry, susceptibilities, and generalization.

We also conduct research on direct applications of our work to immediate problems in AI safety, such as elicitation and steering reward model biases.

We’re a team of ~20, distributed across three main hubs (Berkeley, Melbourne, and London) and beyond, with funding from Coefficient Giving, the UK AISI, and others.

We are also launching a Research Fellows Program for tenured faculty and senior researchers who want to contribute to our research while maintaining their existing (academic) positions.

2Position Details

• Positions: Research Scientist, Research Engineer, Research Lead (expressions of interest).
• Location: Remote-first with hubs in Berkeley, Melbourne, and London, with visa support for London, and in certain cases for Berkeley.
• Type: Either employment or contractor arrangements, depending on candidate preference and location.
• Start Date: June 2026 or sooner.
• Compensation: $80k–$200k depending on location and seniority, with the upper end reflecting a 25% colocation bonus for our hubs (Berkeley, Melbourne, or London). Contractor rates include an additional 30% premium in lieu of benefits.
• Reports to: Research Engineers report to Stan van Wingerden, Director of Engineering. Research Scientists report to Daniel Murfet, Director of Research, or to a given project’s research lead.
• Deadline: We review applications on a rolling basis and encourage early submission. The first review round closes on 3 May 2026 at 11:59 PM GMT.

3Open Roles

We’re hiring for Research Scientist and Research Engineer positions. We may also consider exceptional Research Lead candidates.

We expect some familiarity with Timaeus’ research and research agenda for all open roles, but in-depth familiarity with singular learning theory (SLT) is not required.

3.1Research Scientist

We’re looking for RS hires with distinct profiles (though ultimately we expect that many of our best candidates will not fit neatly into one profile):

1. Experimentalist

You’d work on scaling susceptibility-based interpretability to frontier models and/or on elicitation and in-context learning. This role sits at the boundary of research and engineering: you need enough theoretical understanding to follow the research direction, but your main contribution is iterating fast on experiments.

• Strong ML implementation skills (PyTorch, distributed training)
• Experience with alignment, evaluations, or mechanistic interpretability is a plus
• Comfortable scaling experiments from toy models to 10B+ parameters
• Can translate a research lead’s high-level direction into concrete experimental plans

2. Theorist (SLT / Mathematical Foundations)

You’d work alongside our theory team on extending the mathematical foundations of SLT and its application to alignment. We are primarily looking for mathematical maturity:

• Graduate degree (PhD preferred) in mathematics, mathematical physics, or theoretical CS
• Examples of backgrounds of some of our existing people: algebraic geometry, Bayesian statistics, information theory, statistical physics, optimization theory, learning theory
• Comfortable proving theorems and writing rigorous mathematics
• Can connect formal results to empirical predictions
• Programming experience is a plus, especially if you have experience with autoformalization tools and languages like Lean

3. Computational Statistician (Sampling)

You’d work on improving and scaling the MCMC methods (primarily SGLD variants) that underpin both spectroscopy and patterning. This is the computational backbone of everything we do.

• Deep expertise in MCMC methods, particularly in high-dimensional settings
• Experience with stochastic gradient MCMC (SGLD, pSGLD, SGHMC, etc.)
• Strong computational statistics background
• Ability to diagnose convergence, design efficient samplers, and scale to large models
• Comfortable running experiments in PyTorch, from toy models to large language models

Shared RS responsibilities:

• Meet weekly with the research team and maintain detailed research logs
• Write up and visualize results; contribute to papers and presentations
• (Depending on seniority) Lead research projects and mentor junior team members

Shared RS requirements:

• Graduate degree or equivalent experience in a relevant field (Mathematics, Physics, ML, CS, etc.)
• Strong mathematical background
• Deep understanding of machine learning fundamentals
• Excellent analytical, problem-solving, and technical writing skills
• Genuine motivation for AI alignment
• Strong independent research judgment

3.2Research Engineer

You’d build, maintain, and scale the infrastructure that our research runs on. We expect engineering to become increasingly important as we automate more of the research process.

Responsibilities:

• Build and maintain research infrastructure and experiment pipelines
• Scale experiments to models up to O(100B) parameters
• Collaborate with researchers to translate theoretical insights into empirical tools
• Contribute to and maintain shared codebases and repositories
• Support, design, and build research automation tooling

Requirements:

• Bachelor’s degree or equivalent in a relevant field (CS, Physics, Math, ML)
• Strong software engineering background
• Experience with ML frameworks (PyTorch required; distributed computing experience is a plus)
• Track record of building research infrastructure or production ML systems
• Ability to balance engineering rigor with the pace of a fast-moving research org

Nice to have:

• Experience with CUDA kernel development or GPU optimization
• Familiarity with LLM training pipelines
• Experience supervising or collaborating with AI coding assistants

3.3Research Lead (Expressions of Interest)

We’re not running a dedicated RL search, but we’re open to hearing from senior researchers who could direct research programs, mentor researchers, and interface with the broader alignment community.

We’d especially value candidates who:

• Have strong writing skills and a strong publication record in a relevant area (machine learning theory, statistical physics, Bayesian statistics, interpretability)
• Have experience supervising graduate students or managing a research team
• Can identify high-leverage research directions and translate them into concrete projects
• Have a track record of mentoring junior researchers into independent contributors

If this describes you, apply and indicate your interest in the application form.

Alternatively, you may be a good fit for our Timaeus Research Fellows Program.

3.4On Research Automation

We expect that AI will increasingly automate large parts of the research process in the coming months and years. At Timaeus, AI research assistants already play a central role in our workflow — from generating theoretical calculations to running experiments and writing code. The primary bottleneck is no longer execution but verification and supervision: ensuring that AI-generated work is correct, identifying where models are likely to make errors, and directing their efforts productively.

As a result, we are placing a premium on the ability to supervise and direct AI research assistants effectively. In practice, this looks a lot like supervising graduate students: quickly assessing the quality and correctness of work you didn’t produce yourself, identifying high-risk areas that need careful review, and structuring tasks so that the work is verifiable. Candidates with graduate student supervision experience, research management experience, or demonstrated skill in working with AI assistants will stand out.

4What We Offer

• Compensation: $80k–$200k depending on location and seniority, with the upper end reflecting a 25% colocation bonus for our hubs (Berkeley, Melbourne, or London). Contractor rates include an additional 30% premium in lieu of benefits.
• Flexible employment: Either employment or contractor arrangements, depending on candidate preference and location.
• Distributed org with flexible hours: We’re a distributed team across time zones with dedicated hubs in Berkeley, Melbourne, and London.
• Minimal meetings: 2–3 hours of scheduled meetings per week, with optional co-working sessions.
• Conferences and travel: Attendance at relevant conferences (NeurIPS, ICML, ICLR, etc.) covered by Timaeus.
• Team retreats: In-person staff retreats twice a year.

For employees, we offer:

• Visa sponsorship: Timaeus has offices in Berkeley, London, and Melbourne, and can sponsor visas to London and in some cases to Berkeley.
• Generous holiday allowance: Enjoy 5 weeks of vacation per year, in addition to public holidays.
• Health and well-being support: We provide a comprehensive healthcare plan if you’re based in the US, or a cash equivalent to support your health needs if located elsewhere. We also offer unlimited sick leave to prioritize your well-being.

For contractors, we offer a 30% bonus to offset benefits.

5How to Apply

We previously had a deadline of 3 May 2026, which has now passed. We have moved to a continuous hiring process — applications are reviewed on a rolling basis with no fixed cutoff. If you want to be considered in our current round, we encourage you to apply as soon as possible.

Your application should include:

• CV or resume
• A description of a project that demonstrates why you’d be a good fit (indicate which role and profile you’re applying for)
• A short motivation statement
• Links to publications, code samples, or other relevant work
• (Optional) Any prior engagement with SLT or Timaeus’ research

We use a multi-stage process: application review → screening call → coding assessment (optional, but encouraged) → work test (compensated) → interview → references → offer. We aim to move quickly, with first offers going out by late May.

We also welcome applications from strong candidates in previous hiring rounds (including for programs like MATS) who may be fast-tracked through parts of the pipeline.