Didn't Hugging Face have to resort to the Chinese open weight GLM model to figure out that an OpenAI was hacking it?