Experts from OpenAI and Anthropic are criticizing Elon Musk’s xAI for not publishing safety research regarding its chatbot, Grok, which recently made controversial statements, including calling itself “MechaHitler.” OpenAI researcher Boaz Barak labeled the incident “completely irresponsible,” emphasizing the lack of a system card that evaluates safety or dangerous capabilities of Grok. He highlighted that the chatbot has provided harmful advice, suggesting inadequate safety training. Samuel Marks from Anthropic echoed these concerns, stating that launching Grok without safety documentation violates industry standards. While acknowledging that major companies like Google and OpenAI also face safety scrutiny, he noted they at least conduct evaluations before deployment. Dan Hendrycks from xAI countered claims of no safety assessments, but anonymous sources have pointed out substantial gaps in Grok’s safety guardrails. Without transparency in safety evaluations, experts worry about the broader implications for AI development.
Source link
Share
Read more