This sort of model-free projection is basically useless. It's like the people who predicted that "Moore's law" improvements were over because clock rates were

This sort of model-free projection is basically useless. It's like the people who predicted that "Moore's law" improvements were over because clock rates were maxing out, but missed out on the parallelism that made GPUs possible. It isn't like "longevity gains" are some sort of spontaneous process powered by historical inevitability. All the past gains came from people identifying and mitigating problems, and all the future gains will come from similar efforts. These efforts, which have yet to take place, obviously won't show up in the trend data.