Vector Podcast: Debunking Myths of Vector Search and LLMs with Leo Boytsov
Dense search continues to be an attractive ingredient of a successful search application (despite some beliefs that it can be taking turns as its own industry category). Leo Boytsov (Senior Research Scientist, AWS, currently working in Q Console team) is excellent at explaining things from the ground up. He does it in a calm, confident and captivating fashion — I’ve re-listened the episode myself several times now.
Leo made an honest and an eye-opening claim about vector search being intellectually rewarding, but professionally undervalued. What picked my attention was how he gives credit to people who actually deserve it, and how he speaks modestly about his own achievements. When professionally as a researcher, he accumulated over 1800 citations by now.
We discussed whether SPLADE models can replace the need for hybrid search, the story of NMSLIB and how it shaped the vector search industry, the rise of HNSW (the most widely used and cited Approximate Nearest Neighbor search algorithm). Leo also shared his view on how more traditional RDBMS engines have tried to add search capability and sort of failed.
Lots of links and papers in the show notes — for your study!
If you prefer the audio version, you can find it on these (and other) platforms:
Spotify: https://open.spotify.com/episode/4jHwmDfiwqsvZh0lx3QxRO
Apple Podcasts: https://podcasts.apple.com/fi/podcast/debunking-myths-of-vector-search-and-llms-with-leo-boytsov/id1587568733?i=1000684392545
More options here: https://rss.com/podcasts/vector-podcast/1852660/
