LLM Red Team & Benchmark Evaluation Specialist

This is a fully remote position, open to applicants in New York.

📋 Description

• Evaluation of adversarial models

• Test cutting-edge AI models in areas such as coding, machine learning, analysis, and multi-step agentic tasks

• Detect subtle errors, vulnerabilities, edge cases, and outputs that may appear plausible but are misleading

• Explore scenarios where models seem competent but arrive at incorrect or unsupported conclusions

• Create reproducible experiments to identify and confirm model failure modes

• Benchmarking & Challenge Creation

• Transform identified weaknesses in models into rigorous benchmarking tasks

• Design challenges that are technically rigorous while remaining fair and objectively assessable

• Clearly define task requirements, expected results, and evaluation metrics

• Ensure that tasks necessitate genuine reasoning instead of permitting shortcuts or superficial pattern matching

• Failure Analysis & Reporting

• Document findings with solid evidence, methodologies, and reproducible processes

• Clarify the reasons behind model failures and the capabilities or assumptions that led to the errors

• Generate comprehensive technical write-ups for researchers and task creators

• Monitor recurring failure patterns across different models, prompts, and evaluation environments

• Task Enhancement & Collaborative Research

• Collaborate with task authors to address loopholes, grading inconsistencies, and unintended shortcuts

• Review benchmark tasks for clarity, potential for exploitation, and evaluation reliability

• Share insights with researchers and specialists to enhance benchmark coverage

• Engage in iterative calibration, peer review, and task refinement processes

⛳️ Requirements

• Minimum of 1 year of experience in research, research engineering, security, AI evaluation, or a related technical field

• Proven experience in identifying vulnerabilities, edge cases, or failure modes in LLMs or ML systems

• Background in red teaming, adversarial testing, security research, benchmark development, or thorough model evaluation

• Proficient in Python and Git

• Capability to independently develop scripts, probes, and analyses

• Strong understanding of LLM capabilities, limitations, and evaluation methodologies

• Exceptional written communication and technical documentation abilities

• Creativity, precision, and persistence in addressing ambiguous research challenges

• Reliable availability for approximately 35 hours per week

• A master's degree or PhD in a STEM field is highly relevant

🏝️ Benefits

• Opportunity to work on cutting-edge AI technology

• Collaborative and innovative work environment

• Professional development and growth opportunities