Research Engineer / Scientist, Alignment Science

About the Role

You want to build and run elegant and thorough machine learning experiments to help us understand and steer the behavior of powerful AI systems. You care about making AI helpful, honest, and harmless, and are interested in the ways that this could be challenging in the context of human-level capabilities. You could describe yourself as both a scientist and an engineer.

As a Research Engineer on Alignment Science, you'll contribute to exploratory experimental research on AI safety, with a focus on risks from powerful future systems (like those we would designate as ASL-3 or ASL-4 under our Responsible Scaling Policy), often in collaboration with other teams including Interpretability, Fine-Tuning, and the Frontier Red Team.

Current Topics of Focus

• Scalable Oversight: Developing techniques to keep highly capable models helpful and honest, even as they surpass human-level intelligence in various domains.
• AI Control: Creating methods to ensure advanced AI systems remain safe and harmless in unfamiliar or adversarial scenarios.
• Alignment Stress-testing: Creating model organisms of misalignment to improve our empirical understanding of how alignment failures might arise.
• Automated Alignment Research: Building and aligning a system that can speed up and improve alignment research.
• Alignment Assessments: Understanding and documenting the highest-stakes and most concerning emerging properties of models through pre-deployment alignment and welfare assessments, misalignment-risk safety cases, and coordination with third-party evaluators.
• Safeguards Research: Developing robust defenses against adversarial attacks, comprehensive evaluation frameworks for model safety, and automated systems to detect and mitigate potential risks before deployment.
• Model Welfare: Investigating and addressing potential model welfare, moral status, and related questions.

Representative Projects

• Testing the robustness of our safety techniques by training language models to subvert our safety techniques, and seeing how effective they are at subverting our interventions.
• Run multi-agent reinforcement learning experiments to test out techniques like AI Debate.
• Build tooling to efficiently evaluate the effectiveness of novel LLM-generated jailbreaks.
• Write scripts and prompts to efficiently produce evaluation questions to test models' reasoning abilities in safety-relevant contexts.
• Contribute ideas, figures, and writing to research papers, blog posts, and talks.
• Run experiments that feed into key AI safety efforts at Anthropic, like the design and implementation of our Responsible Scaling Policy.

You May Be a Good Fit If You

• Have significant software, ML, or research engineering experience
• Have some experience contributing to empirical AI research projects
• Have some familiarity with technical AI safety research
• Prefer fast-moving collaborative projects to extensive solo efforts
• Pick up slack, even if it goes outside your job description
• Care about the impacts of AI

Strong Candidates May Also Have

• Experience authoring research papers in machine learning, NLP, or AI safety
• Experience with LLMs
• Experience with reinforcement learning
• Experience with Kubernetes clusters and complex shared codebases

Pay

$350,000 - $500,000 USD annual salary

Logistics

• Minimum education: Bachelor's degree or an equivalent combination of education, training, and/or experience
• Required field of study: A field relevant to the role as demonstrated through coursework, training, or professional experience
• Minimum years of experience: Years of experience required will correlate with the internal job level requirements for the position
• Location-based hybrid policy: Currently, we expect all staff to be in one of our offices at least 25% of the time. However, some roles may require more time in our offices.
• Visa sponsorship: We do sponsor visas! However, we aren't able to successfully sponsor visas for every role and every candidate. But if we make you an offer, we will make every reasonable effort to get you a visa, and we retain an immigration lawyer to help with this.

How We're Different

We believe that the highest-impact AI research will be big science. At Anthropic we work as a single cohesive team on just a few large-scale research efforts. And we value impact — advancing our long-term goals of steerable, trustworthy AI — rather than work on smaller and more specific puzzles. We view AI research as an empirical science, which has as much in common with physics and biology as with traditional efforts in computer science. We're an extremely collaborative group, and we host frequent research discussions to ensure that we are pursuing the highest-impact work at any given time. As such, we greatly value communication skills.

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.