Lao Huang has come to Twitter, and the first article is about a bombshell. He collaborated with multiple organizations to write an article that does not mention KIMI, but rather KIMI. The closed and negative attitude towards Claude between the lines is worth reading. Open Weights and America's AI Leadership In the 1980s, early pioneers of open source software challenged the mainstream notion that software could only progress if companies had strict control over the code. This movement has promoted the establishment of a transparent ecosystem, allowing developers around the world to research, modify, and improve software. Today, the software of the open source community supports most of the foundation of the Internet, and is also the system base relied on by the world's largest technology company, the U.S. military, and federal agencies engaged in key tasks such as scientific research and network security. Open source not only reduces software costs; It has created a shared knowledge foundation on which several generations of American engineers and entrepreneurs have established their own institutional sovereignty. The United States is now facing a similar choice in artificial intelligence. The leadership position of AI in the United States does not depend on any cutting-edge model, but on whether the country can build a strong and open ecosystem that permeates various industries. This is crucial for innovation and prosperity nationwide. It requires expanding the accessibility of AI, encouraging competition, developing powerful application layers, and giving Americans greater control over the technologies they rely on. The open weight model - an AI model that anyone can download, review, modify, and run on their own infrastructure - is an important component of this foundation because it makes advanced AI more accessible, customizable, and widely available. Open weighting has expanded the channels for entering the AI economy. Start up companies, mature enterprises, universities, and public institutions can build on top of advanced models without having to train from scratch or pay the price of cutting-edge models for every task. Open weighting allows each organization to match the right model for the right task at the right cost - leaving cutting-edge capabilities to truly cutting-edge problems and efficient specialized models for other scenarios. This discipline is the key to maintaining economic sustainability as AI expands to billions of daily tasks. The way the United States won the AI era is by infiltrating AI into the workflows of factories, hospitals, farms, classrooms, and small street shops. Open weighting will also strengthen competition, which is the guarantee for AI's benefits to be widely shared rather than concentrated in the hands of a few people. By allowing numerous organizations to build, transform, and deploy advanced models, the competition in open weight manufacturing not only occurs among model developers, but also runs through cloud, chip, application, and service. This competition stimulates innovation, lowers costs, and distributes the benefits of AI widely throughout the economy. Open weighting gives customers greater control. When organizations invest in AI, they want to ensure that they are not locked in the hands of a single supplier, nor do they lose their long-term accumulated knowledge and capabilities. The open weight model provides this guarantee: organizations can control their own data, evaluate and transform models on demand, and deploy them wherever business needs. And when organizations use AI to create value, open weights allow them to truly possess these values through self improving models, proprietary capabilities, and accumulated knowledge - these are exactly what drive American sovereignty and prosperity. Indeed, open weighting carries real and unique risks. Once released, the weight is out of the control of the original developer, and the modified version is difficult to track or retract. But the correct response to this risk is not to prohibit the opening of weights. In a world where cyber attackers are also using advanced AI, defenders need models with comparable capabilities to detect, simulate, and respond to emerging threats. The open model expands defense capabilities, increases transparency, and allows vulnerabilities to be discovered and fixed by numerous teams. In fact, openness may be one of the most important paths towards AI security. Relying solely on closed models is not inherently secure: they may be breached, abused, or fail in ways that are imperceptible to the outside world. Concentrating advanced AI capabilities behind a few closed models will amplify this risk - creating a few single points of failure, weakening competition, and leaving key technologies in the hands of a few suppliers. Open weight models enable a wide community of researchers and developers to review model behavior, identify vulnerabilities, develop defenses, and continuously improve. Just as open source software has proven that transparency can be more secure than obscurity, AI security may also depend on empowering more people to test and reinforce the models that society relies on. It supports rigorous benchmarking and evaluation, red team exercises, and protective measures based on real, proven hazards rather than the assumption that 'closed default is safer'. A powerful AI ecosystem is not destined to emerge. Policy makers now have an important window of action: expanding the supply of computing power for startups and researchers; Invest in shared training assets (datasets, tools, evaluation frameworks); Avoid imposing premature restrictions on open models - which could stifle competition or force innovation overseas - to maintain cutting-edge diversity. These measures should also consider how to leverage a powerful application layer to expand the sovereign use of AI throughout the economy. When shaping this ecosystem, policy makers should be careful not to confuse legitimate model development techniques with invasive behavior. Distillation - using the output of one model to help train or improve another model - is a widely used technique in model improvement, evaluation, and validation. It continues the long-standing tradition of learning, borrowing, and improving 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 is indeed a matter of concern. But these concerns should be addressed through targeted legal and business frameworks, rather than imposing one size fits all restrictions on technologies that play an important role in AI innovation. The AI era can be a prosperous era. Making the right choice, open weighted AI can expand opportunities, strengthen competition, maintain America's technological leadership, mitigate risks, and ensure that the benefits of this extraordinary technology are widely shared throughout the economy. This future is worth building, and the United States should lead its construction. Signatory (25): American Innovators Network· Andreessen Horowitz（a16z）· Arcee AI · Arena · Black Forest Labs · Box · CrowdStrike · Dell · Emergence Capital · Hugging Face · IBM · Linux Foundation · Mariana Minerals · Meta · Microsoft · Mistral · Mozilla · NVIDIA· Palantir · Perplexity · Reflection · Replit · ServiceNow · Telnyx · Y Combinator