Summary & Key Takeaways ​

• The article emphasizes the value of open-source embedding models.
• Proprietary, hosted-only models pose long-term risks due to potential vendor changes.
• Re-calculating millions of embeddings for new models can be costly and time-consuming.
• Open-source alternatives provide flexibility and reduce vendor lock-in.
• The author prefers hosted open-source models, with the option to self-host if needed.

Our Commentary ​

Simon Willison hits the nail on the head here. The long-term cost and effort of re-embedding content when a proprietary model changes or disappears is a massive headache. Open-source models, even if hosted by a third party, offer that crucial escape hatch. This is a practical, developer-centric perspective that we wholeheartedly agree with.