https://venturebeat.com/ai/why-anthropic-and-openai-are-obsessed-with-securing-llm-model-weights/