Why DeepSeek’s sequel failed to impress
DeepSeek's new AI model, v4, was released on April 24th, but failed to impress despite its high performance and low cost. The model's development was likely hindered by increased competition and state meddling in China's AI scene.
DeepSeek, a Chinese AI startup, released its new model, v4, on April 24th. The model's performance is comparable to leading American models from three to six months ago, and it is offered at a significantly lower price. However, the development of v4 was likely more expensive than its predecessor, as the company omitted estimates of training costs in its technical white paper. China's AI scene is becoming increasingly crowded, with competitors like Moonshot, Z.ai, and Alibaba's Qwen models. The country's internet giants are focusing on building AI-powered 'super apps' rather than just developing advanced models. DeepSeek faced challenges, including state meddling, as China's government promotes domestic chips and restricts foreign acquisitions. The company reportedly tried to use Huawei chips to train v4 but ultimately relied on Nvidia's chips, adding cost and time.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.