Embeddings explained on a whiteboard: every text becomes a coordinate

Transcript

like 3D space, not 3D but 100D space where every text of yours gets a coordinate like hello is here, hey is here, I am here and Sanskar Diwari is here what is left in that is that when you search you can see what similarity is so whatever is passed in it, there is similarity so even if your meaning, if your language is totally different you are in a different language they will be much more closer than I am Sanskar Diwari to the hey, its coordinates so basically how does the embedding works you convert that big piece of text into chunks then you convert each chunk into a coordinate which will be in your space when your coordinates come, then what will happen whenever a new query comes what will you do with the new query, you will make a coordinate and you will see who is making it near that coordinate you will make a coordinate and generate embedding and to generate embedding is literally an API call to open okay, we generated embedding and now you will see that in its coordinates, take 5 take 5 out of it now you have saved 5 in the database that this is the actual text of this vector we have taken it out of chunks, this is actually text so did you get the text of 5? give the text of 5 to AI give the text of the user to AI and say based on the question, answer based on the information provided