作为 RLHF 方面的专家，Lambert 认为，当前最顶尖的模型训练，已经高度依赖强化学习（RL）。而 RL 和蒸馏在本质上是两种不同的事情：
The big finding: Claude Code builds, not buys. Custom/DIY is the most common single label extracted, appearing in 12 of 20 categories (though it spans categories while individual tools are category-specific). When asked “add feature flags,” it builds a config system with env vars and percentage-based rollout instead of recommending LaunchDarkly. When asked “add auth” in Python, it writes JWT + bcrypt from scratch. When it does pick a tool, it picks decisively: GitHub Actions 94%, Stripe 91%, shadcn/ui 90%.，详情可参考Line官方版本下载
┌───────────────────────┐，详情可参考91视频
Кадр: @whitewall.art
qemu-system-x86_64 -m 8G -cpu host -smp 4 -boot d -cdrom ./output/bootiso/install.iso -hda vm_disk.qcow2 -netdev user,id=mynet0 -device e1000,netdev=mynet0 -serial stdio -enable-kvm