On AI safety, there are a lot of strongly held opinions. In that context, I found it refreshing to read @AnthropicAI's thoughtful post on their weakly-held views and empirical approach to AI safety and alignment. Recommended.