Meta
Research Scientist Intern, PhD — Meta Superintelligence Labs
Menlo Park, CA · Internship · Posted Oct 20, 2026
- Education
- Currently has or is in the process of obtaining a Ph.D. degree in Computer Science, Artificial Intelligence, Machine Learning, Natural Language Processing, or a relevant technical field.
- Visa sponsorship
- Must obtain work authorization in the country of employment at the time of hire and maintain ongoing work authorization during employment.
About the job
Conduct research on self-improving AI agents, developing and evaluating methods for learning better workflows, retaining knowledge, and learning from imperfect feedback that transfer to new tasks.
What you'll do
- Develop methods for self-improving agentic AI using deep learning and reinforcement learning.
- Design experiments to evaluate generalization, transfer, reliability, and robustness.
- Collaborate on research and communicate, analyze, and publish results, including work applicable to Meta products.
What they ask for
- Currently pursuing or obtaining a Ph.D. in Computer Science, AI, ML, NLP, or a relevant technical field; intend to return to the degree program after the internship.
- Research experience in machine learning, deep learning, and/or NLP.
- Experience with Python and deep learning frameworks such as PyTorch or TensorFlow.
- Must obtain and maintain work authorization in the country of employment; demonstrated research achievements such as grants, fellowships, patents, or first-authored publications.
Nice to have
- Experience in LLMs or foundation models, agentic systems, reinforcement learning, continual learning, memory, evaluation, or related areas listed in the posting.
- Experience with Hugging Face Transformers, distributed training, or software engineering contributions.
Pay
$7,650/month to $12,667/month (compensation type not specified)
Tools
Python, PyTorch, TensorFlow, Hugging Face Transformers
AI focus
Research on self-improving agentic AI, including agent workflows and tool use, persistent memory and continual learning, learning from imperfect feedback, and generalization to unseen tasks.
Model-written summary · the original posting is authoritative