Thanks for the explanation. When storing embeddings, how does the system determine which phrase or words are similar to each other ? Does it assign weights according to some previous train or knowledge? Also, in the VectorDB , do we have the text associated with embedding or it's just the arrays? If so, the system needs to convert it to text again when retreiving the data?
2:50 I don't understand how text is grouped. Who decides or what decides they are grouped by the fact they are athletes and not their nationality? Or why is everything grouped together that says "Cooper" and why isn't it grouped together with all youtubers, humans, or programmers?
It's a bit of a simplified model. What's not really shown here is that it's not embedded in 3 dimensions, but with thousands of dimensions. Some concepts may be close to each other in some dimensions, but far away from each other in other dimensions. Again, even this is a very simplified mental model.