Just now, Huang Renxun tweeted for the first time: Half Silicon Valley, Supporting Kimi K3 Open Source

Just now, Nvidia founder Huang Renxun posted his first tweet of his life:

The account is newly registered, and the tweets are a bit urgent.

In Lao Huang’s first post, he shared a letter signed by NVIDIA titled ‘Open Weights and American AI Leadership’, explaining why open source models are important.

He stated that artificial intelligence will change every industry, empower every company, and be built by every country. Open source models enhance security and network safety, accelerate innovation and dissemination, and achieve sovereignty. The world needs both cutting-edge closed source models and cutting-edge open source models.

In fact, this is an open letter signed jointly by more than 20 technology giants, startups, and investment institutions, including Nvidia, Microsoft Meta， There are also Hugging Face, Mistral, and YC.

They now seem to be planning to join hands to “confront” closed source giants OpenAI and Anthropic.

Here is the full text translation:

In the 1980s, early pioneers of open source software challenged the prevailing notion that software could only progress if companies strictly controlled the code. This movement promotes the establishment of a transparent ecosystem where developers from around the world can learn, modify, and improve software. Today, software developed by the open source community supports most of the functions of the Internet and provides underlying support for the world’s largest technology company, the US military and federal agencies to carry out scientific research, network security and other key mission systems. The significance of open source goes far beyond reducing software costs; It also creates a shared knowledge foundation, on which generations of American engineers and entrepreneurs have built their institutional autonomy.

Nowadays, the United States is also facing a similar decision in the field of artificial intelligence. The evaluation of our leadership position in the AI field does not depend on any cutting-edge model, but on whether the United States can build a strong and open ecosystem and penetrate it into various fields. This is crucial for creating opportunities for innovation and prosperity nationwide. This requires expanding the scope of AI adoption, encouraging competition, building a strong application layer, and giving the American people greater control over the technology they rely on. Open weight models – AI models that anyone can download, check, modify, and run on their own infrastructure – are an important component of this foundation, as they make advanced AI more accessible, flexible, and widely applicable.

Open weighting has expanded the scope of entry for the AI economy. Start up companies, mature enterprises, universities, and public institutions do not need to train models from scratch, nor do they need to pay high fees for cutting-edge models for each task, and can use advanced models for construction. Open weighting enables each organization to match the appropriate model for the appropriate task at the appropriate cost, retaining the ability to scale at the forefront for truly cutting-edge problems, and running efficient professional models in all other fields. This self-discipline will enable AI to sustainably develop economically, especially as its applications expand to billions of daily tasks. If the United States wants to win the era of artificial intelligence, it must popularize it in the workflow of factories, hospitals, farms, classrooms, and street shops.

Open weighting has expanded the scope of entry for the AI economy. Start up companies, traditional enterprises, universities, and public institutions do not need to train models from scratch, nor do they need to pay high fees for cutting-edge models for each task, and can use advanced models for construction. Open weighting enables each organization to match the appropriate model for the appropriate task at the appropriate cost, retaining the ability to scale at the forefront for truly cutting-edge problems, and running efficient professional models in all other fields. This self-discipline will enable AI to sustainably develop economically, especially as its applications expand to billions of daily tasks. If the United States wants to win the era of artificial intelligence, it must popularize it in the workflow of factories, hospitals, farms, classrooms, and street shops.

Open weighting can also enhance competition, which is the key to ensuring that AI achievements are widely shared rather than concentrated in the hands of a few people. By allowing numerous organizations to build, adjust, and deploy advanced models, open weighting has created competition not only among model developers, but also among cloud chips, applications, and services. This competition can stimulate innovation, reduce costs, and widely benefit the entire economy from the benefits of AI.

Open weighting can also give customers greater control. As organizations invest in AI, they hope to ensure that they are not bound by a single supplier and do not lose the knowledge and capabilities accumulated over time. The open weight model helps to provide this guarantee, allowing organizations to control their own data, evaluate models and adjust them according to their own needs, and deploy them wherever needed according to business needs. In addition, as organizations utilize AI to create value, open weighting also allows them to possess these values through self improving models, professional capabilities, and accumulated knowledge, thereby promoting American sovereignty and prosperity.

Indeed, there are some unique risks associated with open weighting. Once released, these weights are out of the control of the original developers, and modified versions are difficult to track or reverse engineer. But the correct way to deal with this risk is not to prohibit the opening of weights. In the era where cyber security attackers exploit advanced AI technology, defenders need to access models with similar capabilities to detect, simulate, and respond to emerging threats. Open models can expand defense capabilities, improve transparency, and enable multiple teams to discover and fix vulnerabilities.

In fact, openness may be one of the most important ways to achieve artificial intelligence security. Relying solely on closed models is not absolutely secure: they may be breached, abused, or encounter faults that external personnel cannot detect. Concentrating advanced AI capabilities on a few closed models will exacerbate this risk. This will lead to a few single point failures, weaken competition, and keep key technologies in the hands of a few suppliers. On the other hand, open weight models allow a wide range of researchers and developers to examine their behavior, identify vulnerabilities, develop security measures, and continuously improve over time. Just as open source software proves that transparency is more secure than concealment, AI security may depend on enabling more people to test and strengthen the models that society relies on for survival. It allows for rigorous benchmarking and evaluation, red team drills, and protective measures related to actual hazards, rather than assuming that closed systems are inherently safer.

A powerful AI ecosystem is not taken for granted. Policy makers have important opportunities for action. This includes expanding access to computing resources for startups and researchers, investing in shared training resources (datasets, tools, evaluation frameworks), and maintaining diversity in cutting-edge fields by avoiding premature restrictions on open models (which could stifle competition or redirect innovation overseas). These measures must also focus on how to expand the autonomous application of AI in the entire economic sector through a powerful application layer.

When building this ecosystem, policy makers should carefully distinguish between legitimate model development techniques and fraudulent behavior. Model refinement, which utilizes the output results of one model to assist in training or improving another model, is a widely used technique for model improvement, evaluation, and validation. It embodies the long-standing tradition of learning, developing, and improving from existing technologies, which has been driving innovation since the rise of the open source software movement. In contrast, the illegal extraction of value from closed models has raised reasonable concerns. These concerns should be addressed through targeted legal and business frameworks, rather than imposing comprehensive restrictions on technologies that play an important role in AI innovation.

The era of artificial intelligence can be a prosperous one. As long as the right choices are made, open AI can expand opportunities, enhance competition, consolidate America’s technological leadership, reduce risks, and ensure that the benefits of this extraordinary technology benefit all aspects of our economy. This future is worth building, and the United States should lead this process.

The stunning Kimi K3 overseas
The trigger for all of this obviously originated from the latest big model Kimi K3 released and open sourced on July 16th. Last week, the release of Kimi K3 was considered a ‘nuclear bomb level’ event in the tech industry. It is not only the largest scale model currently accessible to the open source community (with a total parameter count of 2.8 T), but also makes breakthrough innovations in the underlying architecture, achieving programming capabilities close to Fable 5.

At present, both the coding package and web version of Kimi K3 have been sold out and limited, indicating the enthusiasm of the domestic and foreign technology communities for the new model. The dark side of the moon indicates that K3 will release the complete model weights before July 27th. Hugging Face now even sets up a countdown page for Kimi’s new model:

The whole world is waiting. Major overseas technology communities have begun to spread the “Deployment and Preparedness Guidelines” for K3.

However, after the launch of Kimi K3, OpenAI and Anthropic felt threatened and began accusing Chinese companies of using Model Distillation technology to steal intellectual property, and lobbying US officials. According to reports, the US Treasury Department and other departments have begun considering measures such as banning some Chinese open-source AI models.

This move has also caused panic among Silicon Valley startups. The Little Tech Association, consisting of nearly 200 companies and investment institutions, has warned the government that if cheap open-source models are banned, startups will be forced to pay high API fees to OpenAI and Anthropic, which will directly destroy a large number of financially limited American startups.

Huang Renxun’s debut on the X platform did not showcase GPUs or release large models. Instead, it was a joint open letter supporting open weight AI, which shows the concerns of Silicon Valley giants about the obstacles faced by open source models. This has pushed the debate over the “open source vs. closed source” path to a climax.

What will this big drama of suppressing open source develop into?