OpenAI GPT-3 Text Embeddings - Really a new state-of-the-art in dense text embeddings?

This week, OpenAI announced an embeddings endpoint (paper) for GPT-3 that allows users to derive dense text embeddings for a given input text at allegedly state-of-the-art performance on several relevant tasks. In this post, I will be reviewing how good these new GPT-3 embeddings really are. Are they really a new state of the art? Dense text embeddings are useful for many tasks, including clustering, topic modeling, deduplication, paraphrase mining and semantic search. As part of my research, I've worked on dense text embeddings since 2019 and released my research as part of the sentence-transformers framework, which provides open & free state-of-the-art text embedding models for many use-cases.