Senior AI Researcher
ActiveFence
שכר לא צויןRamat Gan, IL, Ramat Gan, Israel, מרחוקסניורמשרה מלאה
משרה חיצונית, ההגשה באתר החברהאושר שהמשרה פתוחה לפני 14 שעות
We are seeking a Senior AI Researcher to lead evaluation, red-teaming, and reinforcement learning audits of open-weight models. You will build benchmarks and evaluation pipelines to measure model security against threats such as Indirect Prompt Injection.
מה תעשו
- Run post-training experiments with open-weight models in security-focused reinforcement learning environments.
- Analyze loss curves and rollout traces to identify reward hacking, policy convergence issues, and verifier flaws.
- Audit tasks and multi-turn environments for realism, threat-model accuracy, data distribution, and dataset balance.
- Create evaluation cards covering performance across checkpoints, failure modes, rollout length, and task success rates.
- Integrate Dockerized environments into training frameworks and optimize resets, concurrency, and throughput.
דרישות
- M.S. or Ph.D. in Data Science, Machine Learning, Computer Science, or equivalent practical experience in deep learning.
- Hands-on experience training large-scale open-weight models with RL algorithms such as GRPO or PPO.
- Understanding of LLM vulnerabilities, red-teaming, and defensive alignment against IPI attacks.
- Proficiency in PyTorch, Docker, and distributed training architectures.
- Ability to analyze rollout traces, create deterministic rubrics and verifiers, and debug reward shaping flaws.
יתרון
- Experience with standard RL gym formats such as Harbor.
- Experience evaluating tool-use or web-browser agent workflows.
- Familiarity with evaluating open-weight models such as Llama or Mistral against adversarial workloads.
תנאי סף
- M.S. or Ph.D. in a relevant field, or equivalent practical experience in deep learning
- Hands-on experience training large-scale open-weight models with RL algorithms
Reinforcement LearningGRPOPPOPyTorchDockerLLM securityRed teamingDistributed training