ChatGPT Survey - Performance on NLP datasets

• https://opensamizdat.com/posts/chatgpt_survey/

The gist of this paper (I think, I saw this summary elsewhere) is that larger LLM datasets only slightly outperform smaller datasets.