The AI revolution isn’t just seeing the top AI labs compete amongst each other to create the best models, but it’s seeing entire countries compete too.
Marc Andreessen, co-founder of Netscape and prominent venture capitalist, recently articulated a stark warning about the future of open-source AI. He highlights a critical, often-overlooked aspect of the open-source movement: the values embedded within the training data and the organizations behind these models. His concerns revolve around the potential for AI systems to be subtly, yet profoundly, shaped by the cultural and political ideologies of their creators.
“Open weights are great, but open weights are baked, right? The training is in the weights, and you can’t really undo that,” Andreessen said on the TBPN podcast. “And so, are you being trained by a company or an organization or a set of people with American or Western values, or are you being trained by a company with Chinese values?” he asked.
Andreessen admitted that US-trained models have their flaws too, but said that the values being imparted in the Chinese models were far more insidious. “The American dream models have their weirdness. But the Chinese-trained models score really well on things like Marxism. The Chinese benchmarks literally have line items for Marxism. Deep seek is 100 out of 100 on Marxism,” Andreessen claimed.
Andreessen believes these technologies will intermediate critical aspects of our lives. “This is going to be the technology that’s going to intermediate your legal system, the courts, the education system, the medical system. Your kids are going to be learning from these things their whole lives. Do they have Western values, or do they have CCP values? I think it’s really critical,” he said.
He painted a vivid picture of two possible futures. “Just close your eyes and imagine two states of the world. One is where the entire world runs on American open-source AI, and the other is where the entire world, including the US, runs on all Chinese software. I don’t know, for me, that’s a very straightforward, very important topic and a very straightforward answer.”
The debate into US domination of open-source had begun after Chinese company DeepSeek had released a very competent model late last year. But while the model was cheap and extremely effective, it carried with it the ideological underpinnings of the Chinese communist party — the model, for instance, refused to answer questions about the Tiananmen Square massacre, or acknowledge that Arunachal Pradesh was a part of India.
Andreessen’s perspective raises fundamental questions about the influence of AI on society. He touches upon the importance of maintaining Western values in critical sectors like education and justice, warning that without it, the West risks ceding power to China. Andreessen isn’t the only tech leader to voice this concern — OpenAI CPO Kevil Weil had said last week that OpenAI needed to have the best open-source models because the US needed to be ahead of China in open-source adoption. It remains to be seen how the open-source race shapes up, but most tech leaders seem to believe that the winners will get to imprint their cultural values all over the globe.