One of the most common mistakes people make when evaluating the pace of AI research is to look at progress on one type of task and extrapolate it to all tasks that humans can do. AI progress is extremely vertical-specific. In the past year, verifiable domains and in particular code have shown fast progress, which does not extend to other domains. This is because the main driver of AI capabilities remains, to this day, the memorization and operationalization of past data, which can be generated in unlimited quantity in the case of verifiable domains.