AIML - Senior ML Research Engineer, AFM Safety

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

Safety research and methods for Apple's foundation models, covering large language models with a current focus on agentic safety, model character and persona, and multimodal safety across audio and vision. The team works to keep models safe, fair, and robust at scale while meeting Apple's privacy and quality standards, and partners with product teams to ship those safeguards.

• AI misuse and jailbreaks

What you'd do

• Create safety methods for foundation models, with the most weight on agent, persona, and audio and vision work
• Study how agentic systems that use tools can fail or be attacked, then build datasets and tests to measure those risks
• Help shape the model's character and persona so it acts consistently and safely in different settings
• Design training and evaluation approaches that bake safety in for audio and vision models
• Investigate and strengthen how LLMs are aligned to safe behavior and how well that behavior holds up
• Coordinate with engineers, researchers, and product teams to set requirements and get safeguards shipped

What it asks for

• Solid ML background with hands-on model training and published work at venues like ACL, CVPR, NeurIPS, or ICML, or other clear proof of impact
• Strong coding skills, including working in messy, real-world codebases
• Ability to quickly learn the LLM skills needed for safety post-training if not already experienced
• Strong teamwork and communication skills
• Comfort working with offensive or controversial material
• Degree in computer science, machine learning, or related field, or equivalent experience (preferred)
• Direct experience with LLMs and post-training, including safety alignment (preferred)

Worth knowing

• Exposure to offensive or controversial material

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