Timaeus Research Fellows Program

TLDR: Timaeus is launching a Research Fellows Program for tenured faculty and senior researchers who want to contribute to applications of learning theory to alignment while maintaining their existing academic positions.

About Us

Timaeus’ mission is to empower humanity by making breakthrough scientific progress on alignment. Our research focuses on applications of learning theory to foundational problems within alignment. We have hubs in Berkeley, Melbourne, and London.

The Fellows Program

The Timaeus Research Fellows Program is a structured affiliation for researchers at academic institutions or research organizations who want to collaborate closely with Timaeus on SLT-informed alignment research. It is designed for researchers who bring domain expertise — in mathematics, physics, statistics, machine learning theory, or related fields — and want to apply that expertise to open problems in our research agenda.

Fellows are not interns. We are looking for researchers who can operate at the level of a research lead: identifying important questions, designing research programs, supervising projects, and contributing to publications. Timaeus provides the structure, resources, and community to make that collaboration productive.

What the Fellows Program Includes

• Compensation: Competitive rates commensurate with experience and time commitment, structured as consulting or institutional sub-award depending on your university’s requirements.
• Compute access: Access to Timaeus’s computational infrastructure, including GPU clusters for running experiments on models up to O(100B) parameters and LLM access (e.g. Claude team subscription & Claude Code API credits).
• Research tools: Team access to internal tools, codebases, and collaboration infrastructure.
• Community: Integration into Timaeus’s research team, including weekly research meetings, project channels, and direct collaboration with research scientists and engineers.
• Travel: Support for visits to Timaeus hubs (Berkeley, Melbourne, London) and relevant conferences.
• Co-authorship: Joint publications on research conducted during the fellowship.

Structure

• Time commitment: Flexible, negotiated individually based on your institutional obligations and research goals.
• Duration: Initial commitment of one year, renewable by mutual agreement.
• Location: Remote-first. We welcome fellows from any location, with optional in-person time at our hubs.
• Reporting: Fellows work closely with the Director of Research (Daniel Murfet) and the Executive Director (Jesse Hoogland), and may supervise research scientists and engineers on specific projects.

Who Should Apply

We’re looking for researchers who bring:

• A strong track record in mathematics, theoretical physics, machine learning theory, statistics, or a related quantitative field.
• Interest in applying their expertise to problems in AI alignment, interpretability, or the science of deep learning.
• The ability to operate independently and lead research directions.
• Willingness to engage with learning theory and Timaeus’s existing research program — deep prior familiarity with singular learning theory is welcome but not required.

We are especially interested in researchers whose work connects to:

• Bayesian statistics, learning theory, and information theory: Minimum description length, free energy, model selection, asymptotic theory of statistical estimation.
• Algebraic geometry: Resolution of singularities, real log canonical thresholds, asymptotic analysis of integrals.
• Statistical mechanics and phase transitions: Connections between training dynamics and physical systems, renormalization group methods.
• Interpretability and mechanistic analysis: Circuit-level understanding of neural networks, developmental interpretability, training dynamics.
• Sampling and MCMC methods: SGLD, diffusion-based sampling, connections to learning coefficient estimation.

Current Research Areas

• Spectroscopy: Scaling susceptibility-based interpretability to large language models (up to 20B parameters), developing automated methods for discovering internal structure, and applying susceptibility-based probes to alignment-relevant tasks like detecting misalignment. See Gordon et al. (2026).
• Patterning: Using susceptibilities to control what structures neural networks develop during training — including preventing, removing, inducing, and selecting specific internal structures. Near-term applications include steering reward models away from spurious correlations (e.g., format biases). See Wang & Murfet (2026).
• Elicitation: Developing methods to reliably surface model capabilities and knowledge, with applications to evaluation and oversight.

How to Apply

If you’re interested in becoming a Timaeus fellow, please fill out the application form. There is no deadline — we review expressions of interest on a rolling basis and will reach out as we have capacity to onboard new fellows. The fellowship can be structured to meet the requirements of your home institution — we’re happy to work with your department to find an arrangement that works.

This is a new program and the process is intentionally lightweight. The main constraint on our end is time and capacity rather than a fixed pipeline, so timelines may vary. We’ll be in touch once we’ve had a chance to review your application.

For any questions about the program, please contact jesse@timaeus.co.