onsite
LLM Red-teaming Operation, Analyst
LLM Red-teaming Operation, Analyst
As an LLM Red-teaming Operation Analyst, you will join the LLM Global Data Team to devise and manage red-team strategies for LLMs, coordinate testing, and generate vulnerability reports. This role involves leading research into novel issues related to privacy, fairness, and transparency in LLMs to enhance AI system safety and robustness.
About the role
About The Team
As a core member of our LLM Global Data Team, you'll be at the heart of our operations. Gain first-hand experience in understanding the intricacies of training Large Language Models (LLMs) with diverse data sets, and become a leader building and growing a vibrant team together with us!
Your Role Will Involve
- Work closely with LLM core teams to devise a red-team strategy for LLM and manage internal red-teaming operations.
- Coordinate resources across T&S, Data, and product teams to execute red-team tests and generate comprehensive reports, which include, but are not limited to, vulnerability assessments.
- Work with internal researchers and external experts on monitoring the latest developments and industry best practices on red-teaming and advise on the long term strategy.
- Lead research on novel issues related to privacy and also broader questions of fairness, accountability, and transparency related to LLMs.
- Set the research directions and strategies to make our AI systems safer, more aligned and more robust.
Qualifications
Minimum Qualifications:
- Solid understanding of key risks previously facing ByteDance corporate and products. (Or) extensive experience in LLM (Large Language Models) red-teaming or in red-teaming for other content/products.
- A minimum of three years of work experience, with required skills in LLM Adversarial attacks or Jail-breaking methods.
- A team player who knows how to assert influence appropriately. Excellent coordination and persuasion skills are crucial for success in this role.
- Professional proficiency in English
Preferred Qualifications
- Global education or working experience
- Direct research experience over SOTA AI safety topics such as RLHF, adversarial training, robustness, and more.