Research Scientist, AI for Reasoning

Meta is seeking a Research Scientist to join its Fundamental AI Research (FAIR) organization, focused on making significant advances in LLMs reasoning, using reinforcement learning, synthetic data generation and advanced scaffolding/agentic techniques, partially with a focus on formal and informal maths, to join our Paris site. You will have the opportunity to work with a broad and highly interdisciplinary team of scientists, engineers, and cross-functional partners, and will have access to cutting edge technology, important resources, and research facilities.

• Lead, collaborate, and execute on research that pushes forward the state of the art in reasoning research, with an initial focus on formal and informal mathematical
• Work towards long-term high-stakes research goals, while identifying intermediate milestones
• Directly contribute to experiments, including designing experimental details, implement reusable code, running evaluations, and organizing results
• Contribute to publications and open-sourcing efforts
• Mentor other team members. Play a significant role in healthy cross-functional collaboration

• Currently has or is in the process of obtaining a PhD in the field of Computer Science, Mathematics, or similar quantitative field
• Experience training and evaluating large models on State-of-the-Art codebases and developing new architectures, losses and training recipes
• First-author publications at peer-reviewed AI conferences (e.g. NeurIPS, ICML, ICLR)
• Experience in training, fine-tuning, and/or experimenting with foundation models beyond black-box use
• Experience working with SOTA Reinforcement Learning codebases and familiarity with one or more machine learning frameworks (e.g. pytorch, VERL, …)
• Must obtain work authorization in the country of employment at the time of hire, and maintain ongoing work authorization during employment

• Familiarity with the Lean 4 language and ecosystem and mathematical expertise
• Experience with communicating complex research for public audiences of peers