Love these two papers from Anthropic. They are doing a great job analyzing the internals of a llm. arstechnica.com/ai/2025/0…