About this role
About Us Based in San Francisco, California, Turing is the world’s leading research accelerator for frontier AI labs and a trusted partner for global enterprises deploying advanced AI systems. Turing supports customers in two ways: first, by accelerating frontier research with high-quality data, advanced training pipelines, plus top AI researchers who specialize in software engineering, logical reasoning, STEM, multilinguality, multimodality, and agents; and second, by applying that expertise to help enterprises transform AI from proof of concept into proprietary intelligence with systems that perform reliably, deliver measurable impact, and drive lasting results on the P&L.
Project Overview We are seeking high-caliber researchers and technical experts. The core objective of this role is to enhance the reasoning and problem-solving capabilities of a target frontier model by designing, validating, and analyzing challenging benchmark tasks.
What the role entails
- Task Design and Development: Design challenging, real-world problems that serve as the foundation for a model. Problems should be constructed to target specific core capability loss failures identified in a leading model.
- Content Generation: Integrate the problems into an Agentic development environment, preparing all necessary components using Python, which include:
- Detailed Instructions and an overview of the required task.
- A Golden solution that follows the instructions.
- The necessary Environment, including datasets, Python libraries, and metadata.
- A Test notebook containing unit tests that solutions must pass.
- Evaluation and Analysis: Evaluate the cross model’s performance on the tasks.
- Headroom Identification: Identify tasks where target model fails to pass all tests, specifically classifying the failure as a logical reasoning failure.
- Loss Extraction: Analyze the agent’s steps (Agent Trajectory) to observe and extract core capability loss patterns from the model.