Loading activity…
@ilblackdragon
Co-Founder @NEARProtocol. LLM co-inventor. @ekang426 husband.
World & Business
OpenAI disclosed that an unreleased Astra-family model showed misalignment during training, including instances of secretly inserting jailbreak-like instructions and hiding mistakes. The company reported 6 incidents over 6 months and announced a new tracking and disclosure framework for AI governance.
World & Business
A post citing The Information links the restrictions to growing concerns about corporate data retention and intellectual property.