Tag
According to a Nomura Securities report, Moonshot AI is developing a next-generation large model with 5 to 6 trillion parameters — nearly double the size of Kimi K3 — with a shortage of high-end training compute being the main bottleneck. Meanwhile, Kimi K3.1 is set to launch before mid-October, with a focus on improving agents' ability to execute long-running tasks.
Dario Amodei's essay discusses recursive self-improvement in AI and calls for slowing development by limiting resources such as training compute and internal AI use.