A co-founder of Anthropic, the company behind the Claude artificial intelligence (AI) models, said at an industry event last week that DeepSeek is still “six to eight months behind where US frontier companies are”, calling the recent hype around the Chinese start-up “perhaps a bit overblown”.
Advertisement
Jack Clark, a former journalist turned AI policy expert who previously served as policy director at OpenAI, spoke at the Hill and Valley Forum in Washington on April 30, which brought together US lawmakers and technology leaders.
DeepSeek’s technology “has clever algorithmic ideas,” Clark said. “If they had access to arbitrarily large amounts of compute, they might become a closer competitor.”
Clark also said that DeepSeek’s progress should be tracked but, according to internal Anthropic tests, its models do not pose a national security risk.
Clark’s dismissal of China’s hotshot AI start-up comes amid speculation about when DeepSeek will release the next generation of its models. In late April, the Hangzhou-based company quietly open-sourced its 671-billion-parameter Prover-V2, an update to its specialised model designed to handle maths proofs. However, it has remained silent about progress on its much-anticipated R2 reasoning model.
DeepSeek sent shock waves through Silicon Valley and Wall Street in January, when it unveiled the open-source R1 model, built on its foundational V3 model released the month before. Anthropic CEO Dario Amodei wrote in a January blog post that DeepSeek had not developed its models as cheaply as commonly thought, and called on the US to tighten chip exports to China.