The efficiency improvements in training large language models (LLMs) for text embeddings, as described in the paper, can have significant implications for both costs and environmental impact