Re:Claude Sonnet 4 now supports 1M tokens of context
by cambaceres on Tuesday August 12, 2025 @02:58PM
(#44880422)
Sure, I won’t argue against that. The more complex (and fun) parts of the applications I tend to write myself. The productivity gains are still real though.
Quote possibly you are doing very common things that are often done and thus are in the training set a lot, the parent post is doing something more novel that forces the model to extrapolate, which they suck at.