Yesterday, July 24, 25 American technology companies and organizations (the discordant note is given by the French Mistral) as powerful as Nvidia, Microsoft or Meta published an open letter addressed to the White House that they later shared on their networks. Jensen Huang either Satya NadellaCEOs respectively of Nvidia and Microsoft. All of these entities have a common goal: asking the US government not to restrict open-source AI models. The document warns that these restrictions could “slow down competition or take innovation outside the country”
What do they ask for?. That the US government encourages and protects the open AI ecosystem instead of hindering it and they do so with clear requests:
- No prohibitions or premature restrictions on open models.
- Guarantee access to computing capacity for academic research and startups, not just large companies.
- Investment in shared training resources such as databases, evaluation and security tools or frameworks.
- Legislate based on real and proven harms, not hypothetical risks. That the regulations focus on specific misuse, instead of assuming that what is closed is safer by default.
Why is it important. Because this debate about open or closed models of artificial intelligence will determine who controls AI in the next decade, that is, whether it will be concentrated in a small oligopoly of large companies or in an ecosystem made up of thousands of organizations. This will directly affect the costs of AI adoption for SMEs, the competitive ability of the United States against China or how the benefits of the pie are shared.
Security is mentioned in the letter and is an essential topic: depending exclusively on closed models is not a guarantee of security, since they are subject to failures or attacks that may go unnoticed outside the company. The transparency of open models allows a broader community to identify and fix bugs.
Context. The moment chosen to publish this letter comes shortly after the Chinese startup Moonshot AI launched Kimi K3, a true missile to the waterline of the main American models (attending to the evidence) that brings with it a reality: China can win the AI battle and do so with an open model.
The United States’ response is to kill flies with cannon fire: it considers restricting access to Chinese AI models for reasons of national security and intellectual property. Furthermore, Kimi K3 arrived with an accusation under his arm: the director of technology policy Michael Kratsios accused Moonshot AI having “distilled” Anthropic’s Fable model to develop Kimi K3.
In detail. The 25 signatories are Nvidia, Microsoft, Meta, Palantir, Dell Technologies, IBM, Hugging Face, Mozilla, Mistral, Andreessen Horowitz, Y Combinator, American Innovators Network, Arcee AI, Arena, Black Forest Labs, Box, CrowdStrike, Emergence Capital, The Linux Foundation, Mariana Minerals, Perplexity, Reflection, Replit, ServiceNow and Telnyx. Just as important as who’s on the list is who’s not: OpenAI and Anthropic, both valued at nearly $1 trillion and considering IPOs, don’t appear. There is no Google either. Because this debate is not just about national security, but also about two ways of understanding business and who feels like they are winning.
A section worth focusing on is illegal distillation, a technique that allows knowledge to be transferred from a large and complex model to a smaller and lighter one to achieve similar performance with fewer resources. The signatories acknowledge their concerns, but ask politicians not to confuse legitimate techniques with the theft of intellectual property.
Yes, but. Undoubtedly, opening the models allows for a better distribution of technological power and democratizes access to AI, but it is worth remembering that this list of signatories is not neutral: there are component manufacturing companies such as Nvidia or capital funds that have direct interests in the open models prospering because their business depends on the infrastructure and tools to deploy them.
In Xataka | AI companies are buying tons of old books because they are free of AI Slop. Then they destroy them
In Xataka | More and more US companies are using AI models like Kimi K3. The United States is not happy at all
Cover | Gemini
GIPHY App Key not set. Please check settings