I am beyond excited to publish our first Weaviate Podcast interview in-person at the NeurIPS conference with Omar Khattab from Stanford University!<p>I am beyond grateful to have met Omar! I believe strongly that he is at the forefront of Artificial Intelligence technology, especially with the latest developments in Large Language Models, Retrieval-Augmented Generation, and Vector Databases!<p>Omar is a prolific scientist who has published many groundbreaking works, the latest of which being DSPy! DSPy is also an open-source software project on GitHub, achieving roughly 5,000 stars at the time of this writing! I think this is just scratching the surface of where DSPy will go. I think to reach this potential, the next step is developer advocacy and evangelism work. I will be the first to admit that it took me a couple tries to understand the abstractions of DSPy. The framework marries the concepts of pipeline design (really well explained by the abstractions in LangChain, LlamaIndex, Haystack, or Weaviate modules), with prompt and model tuning. I think Omar did an amazing job of explaining this further in the podcast, so I will stop writing this and encourage you to check out the podcast below haha!<p>Omar also touched on ColBERT and multi-vector retrieval methods. These techniques aim to achieve the benefits of the contextual interaction in cross-encoders, directly in a vector index, without the slow inference of applying a cross encoder of a query and millions of documents. Omar again does an incredible job explaining such a complex topic, stay tuned for more updates from Weaviate on multi-vector support!<p>I really hope you enjoy the podcast! I am beyond grateful to have attended the NeurIPS conference and met so many amazing people!<p>https://www.youtube.com/watch?v=CDung1LnLbY