Research, Finetuning Science

Thinking Machines Lab•San Francisco, CA

8d•$350,000 - $475,000•Onsite

About The Position

Thinking Machines is building AI that extends human will and judgment. They are training frontier models with Inkling, developing Tinker to let people make models their own, and crafting interfaces that broaden human-AI communication. This role focuses on building tools that enable people to make AI their own, customizing models to serve their unique needs, including the ability to train model weights. The role involves working on frontier customization techniques and building the post-training engine, Tinker, with a whole-stack understanding of RL science. Findings will directly shape Tinker's training defaults, API design, and the open-source Tinker Cookbook. The position involves working with internal research teams and contributing to open science for external partners.

Requirements

Bachelor’s degree or equivalent experience in Computer Science, Machine Learning, Physics, Mathematics, or a related discipline with strong theoretical and empirical grounding.
Proficiency in Python and familiarity with deep learning frameworks (e.g., PyTorch, TensorFlow, or JAX).
Comfort debugging distributed training and writing code that scales.
Clarity in communication, an ability to explain complex technical concepts in writing.
Strong interest in our mission to enable custom models.

Nice To Haves

A strong grasp of probability, statistics, and ML fundamentals. Ability to distinguish between real effects, noise, and bugs in experimental data.
Prior experience with RLHF, RLAIF, preference modeling, or reward learning for large models.
Experience managing or analyzing human data collection campaigns or large-scale annotation workflows.
Research or engineering contributions in alignment, data-centric AI, or human-AI collaboration.
Experience with RL training stability techniques for large runs.
PhD in Computer Science, Machine Learning, Physics, Mathematics, or a related discipline with strong theoretical and empirical grounding; or, equivalent industry research experience.

Responsibilities

Advance the science of fine-tuning and frontier post-training techniques.
Contribute to areas like LoRA and parameter-efficient fine-tuning and how to push customization quality, efficiency, and reliability to the frontier.
Ship research into product: inform Tinker's training defaults and primitives, and codify best-practice methods as recipes in the Tinker Cookbook.
Improve the stability, efficiency, and reliability of large-scale fine-tuning and RL runs on Tinker.
Share learnings through papers, technical blog posts, and community contributions.
Contribute to areas like LoRA, parameter-efficient fine-tuning, how things interact with RL and post-training, and how to push customization quality, efficiency, and reliability to the frontier.