DeepSeek Sparks Heated Debate in Silicon Valley 200

Due to the Chinese AI company DeepSeek releasing the open-source version of its reasoning model DeepSeek R1 earlier this week, many people in the tech industry have been discussing the company's accomplishments and their implications for the development of AI.

Venture capitalist Marc Andreessen said that DeepSeek is "one of the most amazing and impressive breakthroughs I've ever seen."

DeepSeek R1 appears to have outperformed OpenAI's O1 model on certain AI benchmarks. The company claims that one of its models cost only $5.6 million to train, compared to the hundreds of millions of dollars spent by American companies to train theirs.

In the face of US sanctions that prohibit the sale of advanced AI chips to Chinese companies. MIT Technology Review wrote that the company's success illustrates how sanctions "have driven startups like DeepSeek to innovate in ways that prioritize efficiency and algorithms." (On the other hand, The Wall Street Journal reported that DeepSeek's Liang Wenfeng recently told the Chinese premier that US export restrictions still pose an impediment.)

Neal Khosla, CEO of Curai, offered a simpler explanation, claiming the company is a "CCP state PSYOP" where "plagiarism is cheap, so the price is low, and they hope everyone shifts to American AI companies to boost competitiveness." (His post has been appended with a community note, pointing out that Khosla did not provide any evidence of plagiarism, and that his father Vinod is an OpenAI investor.)

Meanwhile, journalist Holger Zschaepitz said DeepSeek is "possibly the biggest threat to the US stock market"—if a Chinese company can build cutting-edge models at low cost without access to more advanced chips, then "the cost of the technology is called into question. Because hundreds of billions of dollars in capital are being invested in this industry."

In response, Y Combinator CEO Garry Tan believes DeepSeek's success actually benefits its American competitors. He wrote on X: "If training models becomes faster and easier, then the demand for inference (actually using AI) will grow and accelerate even faster, which secures the supply of compute."

Meta's chief AI scientist Yann LeCun pushed back against viewing DeepSeek's claims from a China-vs-US perspective. Instead, he believes the real lesson is that "open-source models are surpassing proprietary models."

"DeepSeek has benefited from open research and open source (such as PyTorch and Llama from Meta)," LeCun wrote on LinkedIn this week. "They came up with new ideas and built upon the work of others. Because their work is published and open source, everyone can benefit from it."

All the debate seems to be driving users to the product. As of the afternoon of January 26, the DeepSeek AI assistant has become the number one free app on the Apple App Store, ahead of ChatGPT.