Open letters about AI development

In recent weeks, a series of open letters advocating for open weight AI models have emerged, penned by Microsoft and the American AI Leadership coalition. Signed by 235 AI-related companies, including NVIDIA, Amazon, and OpenAI, the letter reassures that open weight models are not inherently unsafe. In fact, they allow a larger community of researchers and developers to scrutinize the models, find weaknesses, and enhance their security over time.

One surprising aspect of these open letters is their support for model distillation, where the output from one model assists in training or improving another model. This technique, a staple of open-source software development, fosters innovation and learning from existing technologies.

However, Anthropic, a prominent AI firm, published its own response three days after the initial letter. CEO Dario Amodei expressed concerns about authoritarian countries developing more powerful AI models and the potential misuse of these technologies for cyberattacks or biological attacks. He called for stringent measures against industrial-scale distillation operations, while also defending the company's stance against banning open weight models.

On July 28th, Pacing the Frontier was published, signed by 1,324 employees from frontier AI companies. Their request to the US government was for support in developing the necessary technical and governance tools to manage AI's rapid development. The letter highlighted the intense competitive pressure and accelerated AI progress, driven by automated AI research.

With 80% of their code generated by Claude Code and OpenAI's Sol model reducing end-to-end serving costs by 20%, the urgency of the situation is becoming increasingly clear.