
Senior Deep Learning Scientist, AI Safety
Posted 3 days ago

Posted 3 days ago
This is a fully remote position, open to applicants in California.
• Utilize both fundamental and applied safety research to create, train, refine, and implement large language architectures that support multimodal agentic reasoning systems and establish guardrails across text, audio, image, video, and tool utilization.
• Enhance post-training and alignment strategies, including instruction tuning, preference optimization, RLHF/RLVR/MOPD, to elevate the safety of multimodal agents.
• Investigate and promote AI safety methodologies throughout pre-training, mid-training, post-training, as well as guardrails and steering mechanisms.
• Spearhead the collection, development, and benchmarking of multimodal safety datasets.
• Guarantee rigorous evaluation of model accuracy, safety, and task-completion success.
• Innovate algorithmic advancements for foundational models that handle diverse data types.
• Participate in safety initiatives via Nemotron Omni platforms, guardrails, and external collaborations.
• Tackle significant safety challenges in large language models and multimodal AI products.
• Master’s degree (or equivalent experience) or PhD in Computer Science, AI, or Applied Mathematics, accompanied by 8+ years of pertinent work experience.
• Exceptional programming proficiency in Python.
• Strong foundation in scalable model development.
• Experience with deep learning frameworks, such as PyTorch.
• Comprehensive understanding of ML/DL techniques and contemporary foundational model architectures, including Transformers and mixture-of-experts models.
• Practical experience with post-training multimodal models tailored for multimodal reasoning.
• Demonstrated capability to manage the entire model development lifecycle, encompassing dataset versioning, experiment tracking, and evaluation pipelines.
• Minimum of 1 year of experience in content safety, ML fairness, AI model security, human-AI interaction, or related fields.
• Strong publication record in prestigious AI and machine learning conferences such as NeurIPS, ICML, ICLR, FAccT, or CVPR.
• Experience in applying deep reinforcement learning techniques to train multimodal agents within complex simulations.
• Proven track record in addressing safety and legal and regulatory compliance issues in multimodal, multilingual, or multi-turn systems.
• Familiarity with dynamic and multi-turn alignment benchmarks.
• Experience in at least one risk domain: CBRNE; persuasion and manipulation; discrimination at scale; mental health; LLM security; frontier risks; agentic safety; or multi-turn safety.
• Eligibility to work in the United States, as specified by the US location.
• Highly competitive salaries.
• Comprehensive benefits package.
• Equity opportunities.
• Benefits for you and your family.
Flare
Helix
Weekday (YC W21)
NVIDIA
Get handpicked remote jobs straight to your inbox weekly.