Interpretability Researcher
EleutherAI
📍Remote (Global) 🕔 Full Time
💰 $100K - $150K USD 🔄 Rolling Applications
EleutherAI is seeking talented and motivated individuals to join our Interpretability team to perform cutting edge research with large language and vision models. We aim to better understand the features learned by today’s deep neural networks, so we can better steer their behavior and inform the public and policy makers about their risks and benefits.
We are interested in hiring people at all levels, from current PhD students looking for a part time job to senior researchers who can manage workflows and research agendas independently. Previous experience with large scale AI systems is encouraged but not required.
Research areas we are working on or plan to work on include:
Concept erasure and editing (Belrose et al., 2023) and machine unlearning (Eldan and Russinovich)
Eliciting latent knowledge from neural activations (Burns et al., 2023, Arazia and Mitchell, 2023)
Understanding where and when model behaviors emerge and how they can be controlled during training (McGrath et al., 2022, Biderman et al., 2023)
Understanding neural network inductive biases (Refinetti, Ingrosso, and Goldt, 2022)
Measuring the extent to which neural network behavior is underspecified by data (D'Armor et al., 2022).
…and much more.
Key Responsibilities
Your exact list of responsibilities will depend on your qualifications and experience, but will include at least three of the following:
Planning and running interpretability experiments, and analyzing the results.
Implement and maintain open source interpretability tools and frameworks, such as concept erasure and machine unlearning.
Reviewing and synthesizing relevant academic literature.
Performing mathematical research into the inductive biases of deep learning from a theoretical perspective (see e.g. Yang 2020, Bowman 2023)
Communicating research findings to the broader AI community through academic writings, presentations, and blog posts.
Onboarding and organizing volunteers from our Discord server to help with research and software engineering work.
You may be a good fit if you:
Have significant software engineering experience.
Have experience contributing to distributed research or managing volunteers.
Are comfortable reading and writing mathematical proofs.
Compensation
Salary range: $100k - $150k for full time employees, commensurate with qualifications and experience.
Comprehensive benefits package including health, dental, and vision insurance, retirement plans, and paid time off.
How to Apply
Please submit the following to contact@eleuther.ai:
A resume highlighting relevant experience.
Three papers that you want to build on and how (or what you would do differently)
Links to any public code repositories, publications, or projects relevant to this position.
EleutherAI is an equal opportunity employer. We celebrate diversity and are committed to creating an inclusive environment for all employees.
