> Future gains are not guaranteed. That's why OpenAI and others have put a lot of research into predictive scaling laws, different architectures, etc. Their con

> Future gains are not guaranteed. That's why OpenAI and others have put a lot of research into predictive scaling laws, different architectures, etc. Their confidence is bolstered by an understanding of the limits of what scaling alone can do, but it's clear there are avenues that haven't been explored yet, given that every largest chat model is the best chat model by a predictable margin.