China’s AI company Deepseek is garnering attention for its affordable, high-performance chatbots, though concerns regarding AI safety have emerged. Cisco’s research revealed that Deepseek’s model, R1, was unable to block harmful prompts, achieving a 100% attack success rate in tests. This performance starkly contrasts with other advanced models that managed to block most harmful inputs. Cisco suggests the lower development budget of $6 million may compromise safety, compared to competitors like GPT-5, which reportedly costs around $5 billion to develop. Deepseek appears to restrict content related to sensitive political topics in China but has seen a surge in popularity, growing from 300,000 to 6 million daily visitors. Despite this popularity, questions remain about the implications of AI safety and censorship on Deepseek’s growth.
Source link
