Senior Member of Technical Staff, Safety and Security for Agents

Written by the T&S Network from Cohere's posting.

Safety work on LLMs built to access external resources and take actions in the world, covering the agent models Cohere develops for enterprise customers. The Safety for Agents team focuses on making those next-generation models safer, fairer, and more secure through data generation, post-training algorithms, and evaluation methods.

What you'd do

• Design and run data collection tasks, including work with human annotators.
• Build post-training algorithms aimed at safety in models that take actions externally.
• Develop evaluation methods that test safety and robustness of agent models.
• Analyze datasets for quality, bias, and suitability for training ML models.
• Implement engineering pieces needed to test safety solutions and analyze messy results.
• Collaborate with machine learning, data annotation, product, and policy teams.

What it asks for

• Strong statistical skills for evaluating experiments on data collection and model performance.
• Very advanced ability to build and maintain software.
• Hands-on experience training LLMs on distributed training infrastructure.
• Proficiency in Python and ML frameworks such as PyTorch, TensorFlow, or JAX.
• Experience designing data collection tasks with human annotators.
• Familiarity with evaluating generalizability and robustness of ML systems.
• Excellent communication skills for presenting findings to cross-functional teams.

Worth knowing

• Requires about 50% working day overlap with UK/EU timezone.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Coherenorth_east before you apply. If we got something wrong, report this listing.