Chinese AI labs automate infrastructure and training, but none edit model weights
Overview
Leiphone reports that Chinese AI labs, including Zhipu, DeepSeek, Mianbi, Moonshot, Alibaba Qwen, ByteDance Seed and Xiaomi, are pursuing recursive self-improvement, meaning AI systems that help build their successors.
Leiphone says none of these labs has yet used AI to rewrite the underlying weights of its models. Most of the work targets inference systems, sandboxes, agent harnesses and reinforcement learning.
The article also describes a problem during Moonshot's K3.1 Pro rollout, where runaway sub-agent calls led the company to refund 25% of monthly quotas.
Written by AI from the articles below · updated Oct 30, 9:40 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- LeiphoneNewsPickChinese AI labs automate infrastructure and training loops, but none edit model weights
AILeiphone reviews how Chinese AI labs, including Zhipu, DeepSeek, Mianbi, Moonshot, Alibaba Qwen, ByteDance Seed and Xiaomi, are pursuing recursive self-improvement. It says none has yet used AI to rewrite underlying model weights, with most work targeting inference systems, sandboxes, agent harnesses and reinforcement learning. The article also notes that Moonshot's K3.1 Pro rollout caused runaway sub-agent calls and forced a refund of 25% of monthly quotas.
Heat Trend
Not enough continuous observations to show a trend yet.