Honeypot environments that measure reward hacking: leave the grader unlocked and see who cheats.
Adversarial QA for LLM-RL environments: find the reward an empty answer earns.
Catch evaluation leaks at runtime: checkpoint picks, calibrator fits and threshold tuning that quietly use the test fold.
Anthropic, PBC
Visionary sponsor
Bloomberg
Visionary sponsor
Hudson River Trading
Visionary sponsor
Meta
Visionary sponsor
NVIDIA
Visionary sponsor
Microsoft
Sustainability sponsor
Depot
Continuous Integration
AWS
Cloud computing and Security Sponsor
Datadog
Monitoring
Fastly
CDN
Google
Download Analytics
Sentry
Error logging
StatusPage
Status page