They're only incremental improvements in the sense that developing LLMs is an incremental improvement on stuff that basically worked in the 60s.