Olga
Megorskaya
CEO & Founder
Toloka
Olga Megorskaya is the founder and CEO of Toloka AI, a global tech company powering AI development, from training to evaluation, by optimally combining machine learning technology and human expertise to deliver the highest quality data. She is an expert in designing and building effective business solutions based on human-AI cooperation, human-in-the-loop and crowdsourcing. Prior to Toloka, Olga played a pivotal role in developing data production infrastructure and efficiently leveraging crowdsourced data labeling for ML-based products across diverse industries like IT, E-Commerce, Telecom, Ridetech, and others at Yandex. Her expertise spans critical tech areas such as Natural Language Processing and Computer Vision. Olga is a co-author of research papers focused on efficient crowdsourcing and quality control. She has also shared her insights at esteemed international conferences, including NeurIPS, ICML, and VLDB. Her work has gained recognition in notable media outlets such as Forbes, VentureBeat, Entrepreneur, Bloomberg, as well as leading US ML/AI publications, cementing her reputation as a knowledgeable authority in real-world ML data applications.
16 April 2024 14:00 - 14:30
Generative frontiers: the quest for high-quality AI training data
We stand at the brink of a new horizon in AI development. High quality data is vital for this era. To deliver this data at scale we in Toloka blend the efficiency of Large Language Models (LLMs) and the nuanced insights of domain experts. This fusion not only addresses the monumental task of defining "good" in the context of AI training data but also showcases innovative strategies to produce high quality data at an unprecedented scale. This talk is a deep dive into data quality in various use-case scenarios such as SFT, RLHF, and model evaluation. We will showcase several examples of our work and discuss how data shapes the next wave of AI innovations.