Everything Bob van Luijt said on any show that made the record, most notable first. Each card names its show and opens the statement there.
Van Luijt: RAG carries less hallucination risk than model fine-tuning
“That is something that is, works better than fine-tuning, for example, because if you fine-tune, then you're still dealing with potential hallucination Fair enough, with RAC that's possible too, but it's like, it's less it's less risky.”
Van Luijt: MongoDB is late to the vector database market
“It's great that they've seen the light too. It's a bit late though, but it's like a, you know, good for them.”
Van Luijt: Harvard study values open-source software at $8 trillion
“They estimated that the value of open source to companies right now is eight trillion dollars, right? And if companies would have to rebuild it themselves, that would three and a half times be that number.”
Van Luijt: RAG was the first unique vector database use case
“And reg was the first unique use case, if you will, that emerged around the ecosystem of vector databases.”
Van Luijt: Current industry RAG implementations are primitive
“The way we, and I, with we, I mean like everybody in this room working on this stuff, are doing RAC right now, is actually pretty primitive, right?”
Van Luijt: Pure vector search alone is often not enough
“Vector search alone is often not enough, right?”
Van Luijt predicts 2024 will be the year of generative feedback loops
“I would not be surprised if this year, twenty-twenty-four, is the year of generative feedback loops.”
Van Luijt: Vector models perform poorly on specific alphanumeric IDs
“Matching on A, B, C, one, two, three is horrible. That's like, the model is super bad at that. Why? Because it was never trained on your on your knowledge base.”
Van Luijt: AI models are now the bottleneck for real-time serving
“The bottleneck is now these models.”
Van Luijt: Vector database value comes from indexing, not basic storage
“The vector embedding is just a data type. So it's just an array of floating point numbers. I mean, you can store that in an old Oracle database, right? So, but the thing where it becomes interesting is the index.”
Van Luijt: Virtually every database now supports vector embeddings
“What we try to start to see now is that basically every database under the sun supports vector embeddings.”
Van Luijt: Most billion-scale vector deployments are just similarity search
“So the most deployments, like the really, the billion scale deployments are very much just similarity search, right?”
Van Luijt: Commercial value sits around open source, not within core technology
“The job to be done, the value to be captured does not per se sit in the open source technology itself. It sits in the layers around that.”
Van Luijt: Language is the most common use case for vector embeddings
“Language is the use case we see the most right now”
Van Luijt: More companies stream data with pre-calculated vector embeddings
“More and more companies start to literally stream the data with pre-calculated vector embeddings.”
Van Luijt: OpenAI recommended Weaviate when deprecating its search endpoint
“OpenAI had a they had a search endpoint that they deprecated. This is pre-ChatGBT, so, but they, so they deprecated that search endpoint, And say, well, you know, you just can buy our embeddings. There are three ways that you can index them and search for them…”
Bob van Luijt
the MAD Podcast 2