Why vector databases are a scam (simon-frey.com)

A recent article critiques the emerging trend of dedicated vector databases, such as Pinecone and Qdrant, positing that using existing databases for vector search is often more practical and effective. The author argues that while vector search technology is valuable, it should be viewed as a feature rather than requiring a unique database solution. Introducing dedicated vector databases can lead to significant challenges, including stability risks, the burden of maintenance, and knowledge silos within teams. The piece encourages developers to leverage their current database systems, which already support vector search capabilities, to reduce complexity and enhance functionality without the headaches of onboarding new technologies. The significance of this perspective lies in its potential to streamline application development in the AI/ML community, particularly for businesses looking to improve functionalities like product recommendations. The article cites widely used databases like PostgreSQL, MongoDB, and Redis, which are capable of handling vector operations effectively. Moreover, it highlights that performance arguments often favor established databases over dedicated solutions, with testing showing that current database implementations can outperform specialized options. The overarching message is a call for practicality: by relying on existing, stable technology, teams can avoid unnecessary complications and focus on innovation.