Tech Lead/Manager, Machine Learning Research Scientist- LLM Evals

Аналитика рынка

Стек

Откликнуться

Загрузите резюме — мы свяжем вас с работодателем напрямую через нашу базу.

Описание

As the leading data and evaluation partner for frontier AI companies, Scale is dedicated to advancing the evaluation and benchmarking of large language models (LLMs). We are building industry-leading LLM evals, setting new standards for model performance assessment. Our mission is to develop rigorous, scalable, and fair evaluation methodologies to drive the next generation of AI capabilities. Our Research teams work with the industry’s leading AI labs to provide high quality data and accelerate progress in GenAI research. As the Tech Lead Manager of the LLM Evals Research team, you will lead a talented team of research scientists and research engineers focused on developing and implementing novel evaluation methodologies, metrics, and benchmarks to assess the capabilities and limitations of our cutting-edge LLMs. This role is critical for designing and executing a roadmap that defines best practices in data driven AI development and will accelerate the next generation of generative AI models in partnership with top foundational model labs. You will: Lead a team of highly effective research scientists and research engineers on LLM evals. Conduct research on the effectiveness and limitations of existing LLM evaluation techniques. Design and develop novel evaluation benchmarks for large language models, covering areas such as instruction following, factuality, robustness, and fairness. Communicate, collaborate, and build relationships with clients and peer teams to facilitate cross-functional projects. Collaborate with internal teams and external partners to refine metrics and create standardized evaluation protocols. Implement scalable and reproducible evaluation pipelines using modern ML frameworks. Publish research

Контакты работодателя (email/phone/telegram) скрыты из публичного превью — отправьте резюме, чтобы мы связали вас напрямую.