درباره این اپیزود
Toloka is an AI data platform that produces training data, evaluations, and human expert contributions to help AI labs and enterprises build higher quality models. In a recent episode of Front Lines, we sat down with Olga Megorskaya, Founder and CEO of Toloka, to learn how the company has navigated every major phase of AI development and what she actually believes about the future of human expertise in AI systems.
Topics Discussed:
- How Toloka has evolved from basic crowdsourcing to RLHF to RL gyms across a decade of AI development
- Why the shift from foundational models to post-training and fine-tuning is driving new demand for human data
- How the types of human expertise Toloka requires have shifted as AI capabilities grew
- Why managing human efforts at scale is, in Olga view, a purely technological problem
- The customer concentration risk that came with the era of foundational models
- Why enterprise AI adoption is Toloka main bet for diversifying beyond frontier labs
- Why generating AI output is now easy and why verifying it remains the hard problem
- How physical AI and robotics is creating a new demand curve for human data
- Whether expert recruitment has gotten harder or easier as AI data work has become more respected
- The statistical methods from 1979 that Toloka quality management systems are still built on
