Tool use does not replace larger language models

The author argues that although tool use (web browsing, code execution) allows smaller models—such as a 1-billion-parameter “cognitive core”—to solve many tasks, this is not sufficient: fast, natural, and reliable responses require internal knowledge. As an empirical example the author cites learning badminton; he suggests that even 1 trillion parameters (1T) may not be enough, so demand for larger models will persist in the long term (reference: “Bitter lesson”).