AIML - Machine Learning Engineer, Red Teaming - Responsible AI and Safety

Summary

Would you like to play a part in building the next generation of generative AI applications at Apple? We’re looking for Machine Learning Engineers to work on ambitious projects that will impact the future of Apple, our products, and the broader world. This role is directed at assessing, quantifying, and improving the safety and inclusivity of Apple’s Generative-AI powered features and products. In this role you’ll have the opportunity to tackle innovative problems in machine learning, particularly focused on large language models for text generation, diffusion models for image generation, and mixed model systems for multimodal applications. As a member of Apple’s Responsible AI group you will be working on a wide array of new features and research in the generative AI space. Our team is currently interested in large generative models for vision and language, with particular interest on Responsible AI, safety, fairness, robustness, explainability, and uncertainty in models.

Description

This role focuses on developing, carrying-out, interpreting, and communicating pre- and post-ship evaluations of the safety of Apple Intelligence features. Both human grading and model-based auto-grading are thoughtfully leveraged to power these evaluations. Additionally, this role researches and develops auto-grading methodology & infrastructure to benefit ongoing and future Apple Intelligence safety evaluations.Producing safety evaluations that uphold Apple’s Responsible AI values requires thoughtful data sampling, creation, and curation for evaluation datasets; high quality, detailed annotations and careful auto-grading to assess feature performance; and mindful analysis to understand what the evaluation means for the user experience. This role heavily draws on applied data science, scientific investigation and interpretation, cross-functional communication and collaboration, and metrics reporting and presentation.

Key Responsibilities

• Develop metrics for evaluation of safety and fairness risks inherent to generative AI features.
• Design datasets, identify data needs, and work on creative solutions, scaling and expanding data coverage through human and synthetic generation methods.
• Develop auto-grading technologies and approaches for application in safety evaluations of generative AI features.
• Provide technical direction and expertise to team-wide initiatives in safety auto-grading.
• Use and implement data pipelines, and collaborate cross-functionally to execute end-to-end safety evaluations.
• Work with highly-sensitive content with exposure to offensive and controversial content.

Minimum Qualifications

• MS, or PhD in Computer Science, Machine Learning, Statistics, or related fields; or an equivalent qualification acquired through other avenues.
• Experience working with generative models for evaluation and/or product development, and up-to-date knowledge of common challenges and failures.
• Strong engineering skills and experience in writing production-quality code in Python.
• Deep experience in foundation model-based AI programming (i.e.: using DSPy for optimizing foundation model prompts, for example) and a drive to innovate in this space.
• Experience working with noisy, crowd-based data labels and human evaluations.

Preferred Qualifications

• Experience working in the Responsible AI space.
• Prior scientific research and publication experience.
• Strong organizational and operational skills working with large, multi-functional, and diverse teams.
• Curiosity about fairness and bias in generative AI systems, and a strong desire to help make the technology more equitable.