About this role
This role addresses the need to advance the capabilities of frontier AI systems by focusing on the critical areas of evaluation, data generation, reliability, and agentic science within STEM domains. The team at Turing works to bridge the gap between frontier research and enterprise deployment by creating datasets, environments, and benchmarks that improve model performance in complex reasoning tasks. This position is designed for a research-first individual who thrives on defining new research agendas where established methodologies are lacking. The successful candidate will be instrumental in identifying knowledge gaps and designing rigorous experimental programs to address them.
The scientist will concentrate on several key areas: developing novel benchmarks for STEM, creating methods for high-quality synthetic data generation, and investigating issues related to model hallucination and reliability. Furthermore, the role involves researching agentic science—building systems that can perform extended scientific work—and exploring new research directions in reasoning and AI-for-science. This is a chance to influence the direction of frontier AI research by translating complex problems into actionable, well-defined research projects.
Candidates who excel in formulating precise problems, designing rigorous research programs, and executing experiments rapidly will be well-suited for this function. The ideal candidate possesses a strong foundation in a technical discipline and the ability to navigate the rapidly evolving landscape of large language models.
What you will do
- Identify significant gaps within existing benchmark and evaluation literature for STEM fields.
- Design and construct novel benchmarks across STEM disciplines and general model functionality.
- Develop evaluation methodologies for emerging model capabilities that static benchmarks do not capture.
- Create rigorous methodologies for task generation, data grading, and validation.
- Develop methods for generating high-quality synthetic STEM training data.
- Investigate the impact of data quality on downstream model performance.
- Study hallucination, uncertainty, and reliability in technical domains.
- Research and develop workflows for long-horizon scientific agents.
What you need
- A PhD or equivalent research experience in a technical field such as machine learning, computer science, mathematics, or physics is required.
- Demonstrated ability to formulate and execute original research projects independently.
- A strong comprehension of modern Large Language Models and the current frontier AI research landscape is necessary.
- Excellent skills in experimental design, quantitative reasoning, and scientific judgment are essential.
- The candidate must possess the ability to quickly assimilate unfamiliar technical literature and build expertise in new areas.
- Proficiency in Python is required to build research prototypes and evaluation pipelines.
- Strong technical writing and communication skills are necessary for documenting research findings.
Nice to have
- Strong publication record
Benefits
- $150,000 to $300,000 salary
Skills and tech stack
Questions candidates ask
What is the typical compensation range for this role?
The salary for this role ranges from $150,000 to $300,000.
What is the primary focus of the research?
The research is deliberately focused on frontier STEM evaluation, synthetic data, hallucination and reliability, and agentic science.
What kind of work is involved in agentic science?
This involves researching AI systems capable of performing extended scientific and technical work, including developing workflows involving literature search, coding, and iterative reasoning.
What is the expected pace of the research environment?
Researchers are expected to operate in a fast-moving environment where the research agenda evolves with the frontier.