By: Alex Mercer – SeaPRwire – US AI dominance no longer feels certain. Gary Marcus made that clear in his July 20 blog post. Chinese models have nearly caught up to top American systems. Victory in this race looks impossible. Washington keeps treating AI like a zero-sum game. Marcus calls for a shift. International cooperation and public goods should replace confrontation.
Marcus holds credentials as a prominent AI scholar. He is a professor emeritus at New York University. His work spans machine learning and cognitive science. He pushes for more reliable general AI. His recent piece highlights Moonshot AI’s Kimi K3 model. It matches leading US performance. Open weights allow free downloads and local runs. This development contributed to last week’s market dips in related sectors. It challenges business models at OpenAI and Anthropic.
Earlier models raised similar flags. Zhipu GLM 5.2 and Alibaba’s latest Tongyi Qianwen drew attention. Marcus sees a pattern. Not random events. A clear trend. He predicted this back in early 2025 after DeepSeek’s release. Heavy US focus on large language models would not deliver decisive advantage over China. A draw looked more likely.
His earlier forecasts hold up. OpenAI lacks a strong technical moat and struggles with steady profits. Nvidia faces competition. The CHIPS and Science Act offers limited containment. Models grow cheaper and more efficient. Hallucinations and reliability issues persist. These points have largely materialized.
Marcus criticizes close ties between the US government and Silicon Valley. Visions of generative AI get treated as reality. This leads to strategic missteps. Betting everything on generative AI from the start was a mistake. The field never showed strong enough barriers.
He urges Congress to investigate. Why did the US lose its lead. Whether over-reliance on one technology hurt progress. Intellectual property protection gaps. Immigration restrictions and talent outflow. Chinese AI founders who studied in the US and returned home deserve reflection.
Marcus outlines seven options for the Trump administration. No subsidies. Ban open source. Regulatory moats for US firms. Bailouts for big labs. Full bans on Chinese models. Nationalize OpenAI and Anthropic. He rejects most of them. Regulation to squeeze competitors raises prices and stifles innovation. It hurts American startups too. Government bailouts for loss-making projects make little sense. The industry has not proven sustainable profits yet.
His preferred path rejects winning an AI war. Build something like CERN for AI. International cooperation turns the technology into a global public good. Marcus first suggested this in 2016. Scientists from many countries collaborate on medicine and science goals. Results share worldwide. No monopoly by few nations or companies.
Recent Chinese statements at the World Artificial Intelligence Conference in Shanghai open a window. China supports beneficial and inclusive AI development with all countries. Marcus sees timing for serious consideration. Put AI back in the public domain. International efforts serve medicine and science. This direction holds the most promise now.
A researcher at a European lab shared notes from a recent virtual panel. Participants from several countries discussed model benchmarks. One American engineer admitted surprise at Kimi K3’s accessibility. Local runs changed deployment calculations. Colleagues debated open weights versus closed systems. Reliability concerns surfaced quickly. The conversation moved to cooperation models. Shared datasets for safety research. Joint standards on hallucinations. Practical steps felt more productive than isolation.
Marcus ties this to broader strategy. Talent flows matter. Students who train in the US and build in China highlight policy gaps. Over-betting on one approach narrows options. Cooperation does not mean surrender. It acknowledges current realities. Models advance fast everywhere. Reliability lags behind.
The blog urges Trump directly. A Nobel Peace Prize opportunity exists. Cooperate with China. Direct AI toward public benefit. That gift would serve humanity. Marcus keeps focus on evidence. Performance parity. Market reactions. Prediction accuracy. Policy alternatives.
Teams in AI development should study these arguments. Review internal roadmaps against open-weight progress. Test Kimi K3 and similar models locally. Measure performance on domain tasks. Assess reliability gaps. Factor cooperation scenarios into long-term planning. Governments benefit from independent reviews of talent policies and investment focus. Diversify beyond generative AI. Invest in hybrid approaches that improve trustworthiness. International forums offer venues to test CERN-style pilots. Start small. Medicine imaging. Scientific simulation. Build trust through results.
Author bio: Alex Mercer, seasoned commentator for leading international tech journals with over 15 years covering embedded systems, robotics, and industrial software platforms.