SessionEngineering trackconfirmed
Rethinking Environments for Long Horizon Work
- Day
- Day 2 — Session Day 1 · Tue, Jun 30
- Time
- 11:40am-12:00pm
- Room
- Track 9 · Room 2016
- Track
- Data Quality
Track theme
Data Quality & RL Environments
Data quality for frontier models and reinforcement-learning environments — curation, long-horizon training, and reliability.
Accessible with the Engineering pass and above.
About this session
As autonomous agents push towards longer-horizon tasks, a number of challenges emerge in measuring and improving frontier model capabilities. In this talk, we discuss how long-horizon tasks are defined and measured, how RL environments and verifiers have to scale for more complex and open-ended tasks, and how we navigate these problems at Theta.
