Sam plainly wrote that intelligence of the log of training resources, but that's presumably written in the context of GPT4 style LLMs. The intelligence gains we're seeing right now are not a result of a 100x increase in traditional training resources but rather new ways of training and agentic processes.