Post #412450

Poll: Do you consider embedding models a harmful form of AI? Reposts are welcome! Feel free to elaborate in the replies. Below you can find an explanation of what embedding models are. Embedding models are machine-learning models, that are also trained on ginormous amounts of data, but they are not transformers and don’t produce new output. Instead they condense the input into a large vector that represents its meaning. Because it’s a vector you can do calculations on it. This is useful for stuff like semantic search: Let’s say, you search for "a glass". When there is no direct match, a classical lexical search (e.g n-gram) will give show you "a pair of eyeglasses", whereas semantic search will show you mugs and jars, which are also drinking vessels and thus semantically closer to "a glass" than "a pair of eyeglasses".