kinda question the value of building long term « tracks » over the years in ML venues. A related thought i have is that « canonical » NLP applications (eg summarization) are maybe actually not what users really want or need and were just progressively crystallized over the years

→ View original post on X — @thom_wolf,