Anthropic and the DoW: The Uncertain Future of Open AI Models

A Potential Turning Point for Open AI Models

The current confrontation between Anthropic and the Department of War (DoW) could mark a decisive turning point in how open AI models are perceived and managed globally. As artificial intelligence continues to gain power, established power structures are faced with the necessity of redefining their role in relation to tech companies. Some advocates of open models emphasize the importance of transparency with the adage "not your weights, not your mind." However, this approach raises broader questions as governments become aware of the scale of AI. If this technology is indeed the most influential of our time, why should a single entity, whether a company or a government, control its development and use?

During a weekly discussion on Substack, Nathan Lambert and Dean Ball explored these questions. They concluded that the recent actions of the DoW, particularly the designation of Anthropic as a supply chain risk, could indicate that open models represent a stable solution for power centers over a period of 5 to 10 years.

The Stakes of Open Models

The discussion highlighted several crucial points:

• Why are open models managing to avoid some of the power conflicts observed recently?
• What are the short-term obstacles for open models, and how can they be overcome to strengthen their long-term position?
• What is the balance of capabilities between open and closed models?

It is clear that the need to develop open models is more pressing than ever. However, funding and organizing these initiatives remain major challenges. The strategy of commoditization of add-ons, while effective, shows its limits when the costs of AI models reach astronomical sums, nearing trillions of dollars. The difficulty in monetizing these open models raises the question of who will be able to develop them in the face of significant commercial growth elsewhere in the AI sector.

Key Points from the Interview

• Is the designation of Anthropic as a supply chain risk beneficial or harmful for open models?
• Funding of open models and the growing gap in technological frontiers
• Sovereign AI and global demand for alternatives
• Ecosystem of open models: Qwen, usability, and short-term outlook
• Government power, risk of nationalization, and financialization of computing

What Dean Ball Says

The discussion between Nathan Lambert and Dean Ball starts from the recent news regarding the designation of Anthropic as a supply chain risk. The aim is not to react on the spot to this announcement, but to address its implications for open models. The central question, raised by Nathan Lambert, is whether increased control over AI, with the interest of multiple parties, will be beneficial or harmful to the open weight model ecosystem.

Dean Ball's answer is rather positive: "In the long run, I think this will probably be beneficial for open weight AI." His interest in this issue stems from the importance of governance in advanced AI. Having previously worked on various types of policies, he says he realized that the government would inevitably be involved in managing AI, especially if superintelligence is developed. This technology, he explains, will profoundly influence his daily life and his freedom of expression, and will have implications for national security. The government might not appreciate the use of AI to challenge the status quo, which in his view poses "a major political problem."

Dean Ball sees the issue as complex, as AI involves questions of freedom of expression and national security. The government may be reluctant to allow AI to be used to challenge existing power structures. It will therefore be necessary, he concludes, to "navigate this delicate political landscape."