This 93 pages long paper by Google written by about 128 authors and comes with 160 references. That is a heck of a lot of authors and references when it comes to machine learning & AI papers.
Nevertheless, this paper is disappointing. It scarcely mentions any new methods or advances. It trained on substantially more multilingual language (including e.g. software code) sources than its predecessor. Yes, its capabilities are impressive.
I can not believe that the gravest problems of large language models are misgendering harm and language toxicity. However, that is what this paper suggests.
The 160 references are loaded with leftist and woke ideological desiderata! I already blogged here about it.
No comments:
Post a Comment