Research, RL Scaling

Thinking Machines Lab•San Francisco, CA

4d•Onsite

About The Position

Thinking Machines is hiring a researcher to scale reinforcement learning for frontier models. The role focuses on high-training-compute, long-horizon RL and on jointly designing training recipes and infrastructure. A central area is asynchronous RL, including the algorithmic and systems trade-offs created by decoupling generation from training. The researcher will have full-stack ownership, from algorithms and parallelism plans to monitoring the health of large training runs. Knowledge of inference systems, low-precision numerics, and quantization is recommended to reason quantitatively about rollout throughput and cost and to understand how inference constraints affect training design. The role is based in San Francisco, California. The expected annual salary range is $350,000–$475,000, depending on background, skills, and experience. Thinking Machines sponsors visas and offers generous health, dental, and vision benefits, unlimited PTO, paid parental leave, and relocation support as needed.

Requirements

Proficiency in Python and familiarity with at least one deep learning framework, such as PyTorch, TensorFlow, or JAX.
Comfort debugging distributed training and writing code that scales.
Bachelor’s degree or equivalent experience in Computer Science, Machine Learning, Physics, Mathematics, or a related discipline, with strong theoretical and empirical grounding.
Clear communication skills and the ability to explain complex technical concepts in writing.
Strong research judgment, including clean ablations, honest baselines, and clear technical writing.

Nice To Haves

PhD in Computer Science, Machine Learning, Physics, Mathematics, or a related discipline with strong theoretical and empirical grounding, or equivalent industry research experience.
Strong grounding in reinforcement learning for large language models, such as modern policy optimization methods and their behavior at scale.
Deep understanding of asynchronous RL algorithms, design choices, and trade-offs on both the machine-learning and systems sides.
Experience training large models across many accelerators and familiarity with distributed-stack strategies for parallelism, memory, and communication.
Working knowledge of inference systems and ability to reason quantitatively about rollout-generation throughput and cost.
Experience building or operating decoupled generation/training RL systems at scale.
Experience with RL on verifiable and agentic tasks, including multi-turn environments.
Experience with RL training-stability techniques for large runs.
Familiarity with low-precision training and inference, including numerics, quantization, and their implications for RL.
Hands-on work with LLM serving stacks such as SGLang, vLLM, TokenSpeed, or custom engines.
Experience with scaling studies for large models.
Contributions to open-source training or inference frameworks.

Responsibilities

Co-design RL recipes and the systems that run them, making and validating recipe-level and systems-level choices at frontier scale.
Advance asynchronous RL algorithms.
Improve rollout-generation efficiency and its integration with training, treating inference as a first-class part of the RL loop.
Run frontier-scale RL end to end, including bringing up new models and training setups and keeping large runs stable and healthy.
Jointly optimize RL compute and training efficiency, including accelerator utilization, memory, communication, and low-precision numerics.
Conduct empirical research through ablations and scaling studies, supported by trustworthy instrumentation, and communicate results clearly.