The historical strategy of training language models on publicly available text from websites, books, and other sources has reached a point of diminishing returns, with developers having "largely squeezed as much out of that type of data as they can," according to The Information.
You are viewing a single comment's thread from: