How to apply for Researcher, Interpretability

About EleutherAI

EleutherAI is a nonprofit AI research lab focused on interpretability and alignment of large models. It is known for open source work and public research output. Working there means your research and tools are released to the wider AI community rather than kept internal.

About the role

This role is a research position centered on understanding deep neural network features in large language and vision models. You will plan and run interpretability experiments, build and maintain open source tools such as concept erasure, and communicate findings publicly. The work directly supports the lab's goal of making large models more interpretable and aligned.

A typical day

A typical day likely involves reading recent interpretability papers, writing or debugging experiment code for large models, and analyzing results. You would also spend time maintaining open source tools, discussing findings with collaborators, and preparing public write-ups or talks. Because the role is remote, much of this happens asynchronously, so clear written communication matters.

Who EleutherAI is looking for

• Has hands-on experience training, evaluating, or probing large language and vision models.
• Can design interpretability experiments from hypothesis to analysis, not just run existing code.
• Has contributed to open source interpretability tools or similar research codebases, including maintenance and documentation.
• Can read and synthesize academic literature and explain findings clearly to both technical and non-technical audiences.

Tips for this application

• Show concrete interpretability work: link to papers, blog posts, or repos where you analyzed model features, circuits, or representations.
• Name specific open source tools you have used or contributed to, especially concept erasure or similar interpretability libraries.
• Demonstrate public communication: link to talks, write-ups, or threads where you explained interpretability results to a broad audience.
• Reference EleutherAI's existing interpretability and alignment output and explain how your work connects to or extends it.
• Since the role is remote, briefly note how you have done independent or distributed research work before, including how you plan experiments and share results asynchronously.

What to cover in your cover letter

['A specific interpretability experiment you planned, ran, and analyzed, including what you learned about model features.', 'Your experience with large language and vision models, naming the models or architectures you have worked with.', 'Your contributions to open source interpretability tools, including maintenance, issues, or documentation.', 'How you would communicate findings to the wider AI community, with examples of past public research communication.']

Research before applying

• Read EleutherAI's published interpretability and alignment research, including any work on concept erasure or model features.
• Review the EleutherAI GitHub organization to see which interpretability tools are active and how they are maintained.
• Look at EleutherAI's blog, Discord, and public talks to understand how the lab communicates research.
• Check recent papers or posts from EleutherAI researchers to identify current interpretability questions the lab is pursuing.

Likely interview topics

Based on the job description, expect questions about:

• Walk through an interpretability experiment you designed: hypothesis, method, results, and what you would change.
• How would you approach understanding features in a large language or vision model where ground truth is unclear?
• Describe your experience implementing or maintaining an open source interpretability tool. How did you handle contributions and documentation?
• How do you review and synthesize a body of interpretability literature when the field is moving quickly?
• How would you explain a technical interpretability result to a non-expert audience without losing accuracy?

Common mistakes to avoid

• Applying with only general ML experience and no specific interpretability experiments or feature analysis work.
• Claiming open source experience without links to repos, commits, or maintained tools.
• Writing a cover letter that does not mention EleutherAI's nonprofit, open source, or interpretability focus.