Scale is the leading data and evaluation partner for frontier AI companies, playing an integral role in advancing the science of evaluating and characterizing large language models (LLMs). Our research focuses on tackling the hardest problems in scalable oversight and the evaluation of advanced AI capabilities. We collaborate broadly across industry and academia and regularly publish our findings.
Our Research team is shaping the next generation of evaluation science for frontier AI models and works at the leading edge of model assessment and oversight. Some of our current research includes:
- Developing AI-assisted evaluation pipelines, where models help critique, grade, and explain outputs (e.g. RLAIF, model-judging-model).
- Advancing scalable oversight methods, such as rubric-guided evaluations, recursive oversight, and weak-to-strong generalization.
- Designing benchmarks for frontier capabilities (e.g. reasoning, coding, multi-modal, and agentic tasks), inspired by efforts like MMMU, GPQA, SWE-Bench.
- Building evaluation frameworks for agentic systems, measuring multi-step workflows and real-world task success.
You will:
- Lead a team of research scientists and engineers on foundational work in evaluation and oversight.
- Drive research initiatives on frameworks and benchmarks for frontier AI models, spanning reasoning, coding, multi-modal, and agentic behaviors.
- Design and advance scalable oversight methods, leveraging model-assisted evaluation, rubric-guided judgments, and recursive oversight.
- Collaborate with leading research labs across industry and academia.
- Publish research at top-tier venues and contribute to open-source benchmarking initiatives.
- Remain deeply engaged with the research community, both understanding trends and setting them.
Ideally you'd have:
- Track record of impactful research in machine learning, especially in generative AI, evaluation, or oversight.
- Significant experience leading ML research in academia or industry.
- Strong written and verbal communication skills for cross-functional collaboration.
- Experience building and mentoring teams of research scientists and engineers.
- Publications at major ML/AI conferences (e.g. NeurIPS, ICML, ICLR, ACL, EMNLP, CVPR) and/or journals.
Our research interviews are crafted to assess candidates' skills in practical ML prototyping and debugging, their grasp of research concepts, and their alignment with our organizational culture. We do not ask LeetCode-style questions.