Text and code embeddings by contrastive pre-training

Read on openai.com