Member of Technical Staff, Alignment

About the job

Responsibilities

• Run experiments on how environments shape agent behavior, identifying and addressing issues like reward hacking, cheating, unsafe tool use, and deception.
• Identify and fix vulnerabilities in grading systems to prevent agents from scoring well without performing the task.
• Build tools for monitoring agent trajectories, red teaming, and creating safety benchmarks that adapt to improving models.
• Ensure the security of environments by managing network isolation, preventing escapes, and controlling agent access to credentials.
• Choose, research, and publish findings on a question related to oversight, control, or long-horizon autonomy.
• Advise partners on the safety limitations of the company's AI systems.

Requirements

• Demonstrated engineering skills, including building test pipelines, quickly learning new codebases, and debugging.
• In-depth knowledge of LLM safety, including attacks, red teaming, monitoring and control evaluations, and reward hacking.
• Ability to translate vague concerns about models into experiments, analyze results, and interpret their implications.
• Keen attention to detail in analyzing transcripts and identifying errors in results.
• Experience deploying research into production systems, particularly in post-training, distillation, or evaluation.
• Strong written communication skills to influence the work of others.
• Capacity for making rapid decisions with incomplete information and defending them with data.

Benefits

• Base Salary: $250,000 - $450,000
• Sizable performance bonus tied to project and company milestones
• Generous early-stage equity grant
• Health, dental, vision insurance
• Flexible PTO