> We do actually understand generally well enough what is happening. See the comment on the "Golden Gate Bridge" version of Claude: "The fact that we can find

> We do actually understand generally well enough what is happening. See the comment on the "Golden Gate Bridge" version of Claude: "The fact that we can find and alter these features within Claude makes us more confident that we’re beginning to understand how large language models really work." (emphasis mine) https://www.anthropic.com/news/golden-gate-claude https://www.anthropic.com/news/golden-gate-claude