>Everybody I know that worked in the field of so called AI 8 years ago expected exactly that

Find me any paper or blog from 10 years ago (or anytime before "All you need is attention" dropped) that predicted we'd be where we are with AI now and (critically) the specific means of how we'd get here. All there are are general "Neural nets will get better and better" wishy washy predictions of the type anyone can blurt out. Everything that happened with transformer architectures and their scale properties and their application to large language models, that was as out of the blue as it gets, even for those on the inside of the general neural net field.