I'm so done with these overhyped large language models. We all know they're just aggregations of web scraping and tokenization, yet we pretend they're some kind of breakthrough. Where's the actual innovation?