Study: Noisy data severely degrades RLVR training performance
ddkang · x · 2026-08-28
Research by Daniel Kang's team refutes the claim that RLVR is robust to noisy data. They found prior claims of success with '100% incorrect' data were due to contamination with correct labels. A rigorous curation pipeline shows noisy data degrades test accuracy by over 9%, which algorithmic improvements fail to mitigate.
More from Research
- Anthropic launches MHS standard for AI agents to safely control physical lab equipment — jmuiuc · 2026-08-28
- The Hundred-Page Language Models Book Released — burkov · 2026-08-28
- Bittensor humanoid robot competition yields superhuman long jump skills — markjeffrey · 2026-08-28
- Research shows injecting 1,000 fingerprints into open models causes almost no performance drop — 0xsachi · 2026-08-28
- AI safety researchers concerned multi-agent training may make models more helpful to AIs than humans — dhadfieldmenell · 2026-08-28
- LLMs can't originate a KRAS-style insight, but excel at cross-domain hypothesis synthesis — Ghost_Pilot_MD · 2026-08-28