Keep pulling the thread on Nathan Lambert.
The US government imposed sudden export controls on Anthropic's Mythos and Fable models.
Anthropic blocked access to its Mythos and Fable models in response to US government export controls.
Anthropic's safety measures for its models were inconsistent, applying classifier-based restrictions for some use cases and silent performance degradation for others.
Chinese AI researchers heavily rely on Anthropic's Claude models, often accessing them through proxy services, and are now scrambling due to the prospect of stricter user verification rules.
The US government implemented export controls on Anthropic's models as an indirect method to force the company to shut them down.
The US lacks the necessary technical evaluation frameworks to make informed regulatory decisions about AI models, especially for narrow jailbreaks.
Future AI models will become significantly more intelligent, leading to the emergence of autonomous AI agents that will trigger a new, more intense era of governance challenges.
The US federal government would likely ban a surprisingly strong open-source model released by a Chinese company.
To date, almost all of the most capable open-source "mixture of experts" (MoE) models have been built by Chinese companies.
The vast majority of current language model research is being conducted using Alibaba's Qwen 3 and Qwen 3.5 models.
Top US AI labs have a significant training compute advantage over top Chinese AI labs.
Chinese AI labs are now using Huawei chips for large MoE model inference at a reasonable price point.