logoalt Hacker News

linkregistertoday at 3:05 PM0 repliesview on HN

There's no evidence to suggest that poisoning impacts generative model training in 2026. Frontier labs spend billions on tightly focused training plans, developing assessments and pursuing the long tail of assessment failures.