What else happened on Nov 13th
Almost all of the day's lab posts are about who checks models and how. Anthropic put together advice on building safety evals, and Generality Labs took apart Anthropic's biology benchmark and calculated what share of cheating the benchmark's LLM judge misses. Transformer writes that the US doesn't yet have enough people for a strict regime of ongoing external audits of frontier AI companies, like the one in the Massachusetts bill. Two more posts are about China: views from five Chinese experts on AI talks with the US, and a warning from China's state security service about the intermediaries that users in China go through to connect to foreign models.