The Closing Open-Closed Gap
In issue 465 of the Import AI newsletter, Jack Clark analyzes the rapidly narrowing capability gap between open-weight and closed-source AI models [[33]]. The analysis suggests that open ecosystems are professionalizing quickly, allowing them to iterate and optimize models at a pace that rivals proprietary labs.
The Kimi K3 Milestone
A prime example of this acceleration is the release of Kimi K3, a massive 2.8 trillion parameter model [[37]]. The newsletter highlights how Chinese labs are demonstrating remarkable capital efficiency, turning resources into compute, data, and talent in ways that allow them to catch up to frontier performance without matching the exact spending profiles of US counterparts.
New Frontiers in AI Governance
Beyond model releases, the issue covers Demis Hassabis's new policy proposals for global AI governance and security [[39]]. As open models become more capable, the newsletter argues that policy frameworks must evolve to address the unique security dynamics of widely accessible, high-performance AI systems.