The Interpretability team at Anthropic is working to reverse-engineer how trained models work because we believe that a mechanistic understanding is the most robust way to make advanced systems safe. We’re looking for researchers and engineers to join our efforts.
People mean many different things by "interpretability". We're focused on mechanistic interpretability, which aims to discover how neural network parameters map to meaningful algorithms. If you're unfamiliar with this type of research, you might be interested in this introductory essay, or Zoom In: An Introduction to Circuits. (For a broader overview of work in this space, one of our team's alumni maintains a helpful reading list.)
Some useful analogies might be to think of us as trying to do "biology" or "neuroscience" of neural networks, or as treating neural networks as binary computer programs we're trying to "reverse engineer".
Some of our team's notable publications include A Mathematical Framework for Transformer Circuits, In-context Learning and Induction Heads, and Toy Models of Superposition. This work builds on ideas from members' work prior to Anthropic such as the original circuits thread, Multimodal Neurons, Activation Atlases, and Building Blocks.
We aim to create a solid foundation for mechanistically understanding neural networks and making them safe (see our recent vision post). In the short term, this means a we focus a lot of our attention on the issue of "superposition" (see Toy Models of Superposition, Superposition, Memorization, and Double Descent, and our May 2023 update). But this is just a stepping stone towards our goal of mechanistically understanding neural networks.
About Anthropic
Responsibilities:
- Develop methods for understanding LLMs by reverse engineering algorithms learned in their weights
- Design and run robust experiments, both quickly in toy scenarios and at scale in large models
- Build infrastructure for running experiments and visualizing results
- Work with colleagues to communicate results internally and publicly
You may be a good fit if you:
- Have a strong track record of scientific research (in any field), and have done some work on Interpretability
- Enjoy team science – working collaboratively to make big discoveries
- Are comfortable with messy experimental science. We're inventing the field as we work, and the first textbook is years away
- You view research and engineering as two sides of the same coin. Every team member writes code, designs and runs experiments, and interprets results
- You can clearly articulate and discuss the motivations behind your work, and teach us about what you've learned. You like writing up and communicating your results, even when they're null
Familiarity with python is required for this role.
Annual Salary
- The expected salary range for this position is $250k - $520k
Deadline to apply: We expect to be hiring for this role intermittently, but our plan is to hire a few people in the next 2-3 months, and then will likely slow hiring while the team settles. (This opportunity was originally posted early July.)
US visa sponsorship: We do sponsor visas! However, we aren't able to successfully sponsor visas for every role and every candidate; operations roles are especially difficult to support. But if we make you an offer, we will make every effort to get you into the United States, and we retain an immigration lawyer to help with this.
LinkedIn tracking: #LI-DNI
We encourage you to apply even if you do not believe you meet every single qualification. Not all strong candidates will meet every single qualification as listed. Research shows that people who identify as being from underrepresented groups are more prone to experiencing imposter syndrome and doubting the strength of their candidacy, so we urge you not to exclude yourself prematurely and to submit an application if you're interested in this work. We think AI systems like the ones we're building have enormous social and ethical implications. We think this makes representation even more important, and we strive to include a range of diverse perspectives on our team.
Compensation and Benefits*
Equity - On top of this position's salary (listed above), equity will be a major component of the total compensation. We aim to offer higher-than-average equity compensation for a company of our size, and communicate equity amounts at the time of offer issuance.
US Benefits - The following benefits are for our US-based employees:
- Optional equity donation matching at a 3:1 ratio, up to 50% of your equity grant.
- Comprehensive health, dental, and vision insurance for you and all your dependents.
- 401(k) plan with 4% matching.
- 22 weeks of paid parental leave.
- Unlimited PTO – most staff take between 4-6 weeks each year, sometimes more!
- Stipends for education, home office improvements, commuting, and wellness.
- Fertility benefits via Carrot.
- Daily lunches and snacks in our office.
- Relocation support for those moving to the Bay Area.
UK Benefits - The following benefits are for our UK-based employees:
- Optional equity donation matching at a 3:1 ratio, up to 50% of your equity grant.
- Private health, dental, and vision insurance for you and your dependents.
- Pension contribution (matching 4% of your salary).
- 22 weeks of paid parental leave.
- Unlimited PTO – most staff take between 4-6 weeks each year, sometimes more!
- Health cash plan.
- Life insurance and income protection.
- Daily lunches and snacks in our office.
This compensation and benefits information is based on Anthropic’s good faith estimate for this position as of the date of publication and may be modified in the future. Employees based outside of the UK or US will receive a different benefits package. The level of pay within the range will depend on a variety of job-related factors, including where you place on our internal performance ladders, which is based on factors including past work experience, relevant education, and performance on our interviews or in a work trial.
How we're different
The easiest way to understand our research directions is to read our recent research. This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences.
Come work with us!