Member of Technical Staff, Safety for Agents

Cohere Cohere · AI Frontier · London, United Kingdom · Modeling

Cohere is seeking a Member of Technical Staff for their Safety for Agents team. This role focuses on developing better, fairer, and more trustworthy LLMs by working on data generation, post-training algorithms, and evaluation methods for AI safety. The position involves collaboration with ML, data annotation, product, and policy teams, and requires strong statistical, software engineering, and data analysis skills. The ideal candidate will have experience training LLMs and evaluating their generalizability and robustness, with a publication record in top-tier venues.

What you'd actually do

  1. Your primary focus will be on data generation, post-training algorithms, and evaluation methods to ensure Safety in the next generation of models that can access external resources and take actions in the world.
  2. You will work closely with other cross-functional machine learning teams and data annotation teams, and will also collaborate with product and policy teams.
  3. It will require curiosity to tackle totally new scientific problems, engineering skills to implement the pieces we need to test solutions to these, and a desire to dive into messy data and results.
  4. You will be on a small team with a lot of autonomy and decision-making power, responsible for making the next generation of LLMs better for society as a whole.

Skills

Required

  • Strong statistical skills and experience evaluating scientific experiments related to data collection and model performance.
  • Extremely strong software engineering skills.
  • Strong expertise in designing and conducting data collection tasks, including working with human annotators.
  • Experience analyzing datasets with respect to their quality, biases, and suitability for training ML models.
  • Hands-on experience training large language models (LLMs) on distributed training infrastructures.
  • Familiarity with evaluating and improving the generalizability and robustness of ML systems.
  • Proficiency in programming languages such as Python and ML frameworks (e.g., PyTorch, TensorFlow, JAX).
  • Excellent communication skills to collaborate effectively with cross-functional teams and present findings.

Nice to have

  • One or more papers at top-tier venues (such as NeurIPS, ICML, ICLR, AIStats, MLSys, JMLR, AAAI, Nature, COLING, ACL, EMNLP).

What the JD emphasized

  • primary focus will be on data generation, post-training algorithms, and evaluation methods to ensure Safety
  • ensure Safety in the next generation of models that can access external resources and take actions in the world
  • Strong statistical skills and experience evaluating scientific experiments related to data collection and model performance.
  • Extremely strong software engineering skills.
  • Strong expertise in designing and conducting data collection tasks, including working with human annotators.
  • Experience analyzing datasets with respect to their quality, biases, and suitability for training ML models.
  • Hands-on experience training large language models (LLMs) on distributed training infrastructures.
  • Familiarity with evaluating and improving the generalizability and robustness of ML systems.
  • One or more papers at top-tier venues (such as NeurIPS, ICML, ICLR, AIStats, MLSys, JMLR, AAAI, Nature, COLING, ACL, EMNLP).

Other signals

  • safety
  • responsible AI
  • LLM evaluation
  • data generation
  • post-training algorithms