AI Research Scientist, Language - Monetization GenAI

Meta•Menlo Park, CA

About The Position

We are developing the GenAI models that power Meta's advertising ecosystem at scale. Our team operates at the cutting edge of LLM post-training & applications, creating systems that help millions of advertisers succeed while driving significant revenue gains. Our team works at the intersection of AI and Monetization, and we've launched multiple 0→1 GenAI advertising products. We handle the full LLM lifecycle: post-training, evaluation, deployment, and product launch.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience
Has obtained a PhD in Computer Science, AI/ML, or a relevant technical field
2+ years of experience in NLP or multimodal LLM research and development
Hands-on experience LLM post-training and/or reinforcement learning (reward modeling, PPO/GRPO, RLHF)
Strong experience with ML tech stack (e.g., PyTorch, building data pipeline)
Experience leading major technical initiatives with cross-functional impact, and/or influencing strategy across multiple teams
Demonstrated significant industry influence in the field of AI and/or recently published research in leading peer-reviewed conferences (e.g., ACL, NeurIPS, ICML, ICLR, AAAI, KDD, CVPR, ICCV)

Nice To Haves

Ability to bridge modeling and production - turning novel research ideas into shipped products
First-author publications at top peer-reviewed conferences (e.g., ACL, NeurIPS, ICML, ICLR, AAAI, KDD, CVPR, ICCV)
Background in ads systems
Experience with recommendation systems or ranking models

Responsibilities

Improve content understanding and knowledge for GenAI-powered advertising
Leverage state-of-the-art multimodal LLMs and agentic models to extract, structure, and retrieve valuable signals that power ad creative generation and optimization
Pioneer the use of Reinforcement Learning from Human Feedback (RLHF) to post-train LLMs for real-world advertising performance
Develop RL post-training frameworks that use actual ads data to fine-tune LLMs to generate ad creatives that resonate with users