Mid-Training Researcher (AI)

Location: Hybrid role based in San Francisco, California, United States

Salary: $350,000–$475,000 annual salary

Company

Thinking Machines Lab builds AI systems intended to extend human judgment and communication, including frontier models, model-customization tools, and human-AI interfaces.

What you will do

• Own the sourcing, curation, synthesis, filtering, deduplication, verification, and rewriting of training data for model capabilities and knowledge areas.
• Design and measure knowledge-improvement interventions, including targeted corpora, synthetic rephrasings, question-answering over source documents, and knowledge-dense data mixes.
• Introduce and evaluate behaviors during mid-training and coordinate with post-training researchers on which behaviors belong in mid-training or reinforcement learning.
• Build automatic data-quality systems, including quality classifiers, LLM-based judges, synthetic-data verifiers, and fine-grained data-attribute controls.
• Develop and tune mid-training recipes covering dataset collections, training stages, annealing schedules, and hyperparameters.
• Define meaningful evaluations, debug unexpected training results, measure scaling behavior, and explore new training-data methodologies.

Requirements

• Proficiency in Python and familiarity with deep learning frameworks such as PyTorch, TensorFlow, or JAX.
• Experience 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.
• Strong theoretical and empirical grounding, with the ability to communicate complex technical concepts clearly in writing.
• Ability to work in a hybrid role based in San Francisco, California.

Nice to have

• Strong understanding of probability, statistics, and machine learning fundamentals.
• Experience building training datasets for large models, including synthetic-data pipelines, real-user data, collection, curation, filtering, or mixture design.
• Experience with knowledge injection, continued pre-training, domain adaptation, or measuring model knowledge.
• Experience building model-based data-quality systems, LLM judges, verification pipelines, or rewriting systems at billion-token scale.
• Research or engineering contributions in alignment, data-centric AI, or human-AI collaboration.

Culture & Benefits

• Individual contributor role within the Research Team, reporting to the Head of Research.
• Close collaboration with a small group of post-training researchers.
• Health, dental, and vision benefits.
• Unlimited paid time off and paid parental leave.
• Visa sponsorship and relocation support are available as needed.