Open-Model AI Safety Lead: Red Team & Validation | Equity

Reflection AI; New York (NY); On-site; USD 180,000 - 240,000; Full time

Benefits offered by this job

Top-tier compensation

Comprehensive medical, dental, and vision insurance

Fully paid parental leave

Paid time off

Daily lunch and dinner provided

Job summary

A leading AI research firm in New York, NY, is looking for a specialist to oversee the red-teaming and adversarial evaluation pipeline for their models. The ideal candidate should possess a graduate degree in Computer Science or a related field and have a deep understanding of LLM safety and adversarial techniques. A strong software engineering background is essential, along with the ability to make high-stakes decisions regarding model safety. This role offers top-tier compensation and comprehensive benefits.

Qualifications

• Strong technical understanding of LLM safety and adversarial attacks.
• Experience in building automated evaluation pipelines.
• Ability to make high-stakes decisions regarding model safety.

Responsibilities

• Own the red-teaming evaluation pipeline for models.
• Work with Alignment team on safety findings.
• Validate model releases meet risk thresholds.
• Develop automated safety benchmarks.
• Implement state-of-the-art jailbreaking techniques.

Skills

LLM safety expertise

Software engineering skills

Adversarial attack knowledge

Reinforcement Learning (RLHF/RLAIF) understanding

Education

Graduate degree in Computer Science or related discipline

Job description

A leading AI research firm in New York, NY, is looking for a specialist to oversee the red-teaming and adversarial evaluation pipeline for their models. The ideal candidate should possess a graduate degree in Computer Science or a related field and have a deep understanding of LLM safety and adversarial techniques. A strong software engineering background is essential, along with the ability to make high-stakes decisions regarding model safety. This role offers top-tier compensation and comprehensive benefits.