From 2022 onwards, America Has had a strong lead in artificial intelligence Thanks to the advanced models of high-flying firms like OpenAI Google DeepMind Anthropic and xAI. Experts are increasingly concerned that the US has fallen behind in developing open-weight AI that can be downloaded and adapted locally.
Kimi Z.ai and other Chinese manufacturers offer open models. AlibabaThen, DeepSeek Researchers and engineers around the world are increasingly interested in AI, making America a laggard. “The US needs open models to cement its lead at every level of the AI stack,” Nathan Lambert is the founder of Nathan Lambert. ATOM (American Truly Open Models) ProjectWIRED interviews.
Only the most sophisticated models of US firms can be accessed via a chatbot or sending questions to company servers using an API. OpenAI, Google, and other companies have developed open-weight models. However, they are less powerful than Chinese versions, as the latter are more flexible and provide better developer support. Chinese model makers can also gain from the open-sourcing of their models. The best ideas from researchers outside China are incorporated into new releases.
Lambert, also a research scientist at the Allen Institute for AI, an organization based in Seattle Washington, created the ATOM project in order to raise awareness of the potential risks that come with falling behind the US in the open-source world. Lambert says that the US needs to adopt cutting-edge models. This is because using foreign models could become problematic in case they are suddenly stopped or closed source.
Lambert says that the open model also encourages innovation and experimentation among startups and researchers. In addition, Lambert says that companies can also benefit from open models. sensitive information They need models they can use on their hardware. “Open models are a fundamental piece of AI research, diffusion, and innovation, and the US should play an active role leading rather than following other contributors,” Lambert Says
ATOM Project was launched on 4 July. The project presents compelling arguments for openness, and shows the Chinese open weight models are now more popular than US ones.
Meta, a US-based social media company, was responsible for launching the open source AI revolution. released LlamaThe open-weight Frontier model is scheduled to be introduced in July 2023. Meta at the time saw Llama’s potential to get into AI. Its new model was very quickly popularized by researchers and entrepreneurs.
Meta, and many other US AI firms, have been fixated with the goal of creating superhuman or human-level AI before competitors. This has resulted in a lack of openness. Recently, Zuckerberg re-energized Meta’s efforts in AI with a string of expensive hires A new “superintelligence” lab. Zuckerberg has also indicated Meta is no longer free to open source its top models.
China’s technology industry, on the other hand, has shifted toward more openness in 2018. DeepSeek, then a little-known startup at the time, published an article in January 2025. open model called DeepSeek-R1 The world was astonished by its capabilities, and it cost a fraction as much to train than the major US models. In the years since, many Chinese firms have launched powerful open-weights with additional innovations.
