3 ms·
Maybe not interfacing directly with a LLM ala ChatGPT but as an extension of vector search using word embeddings. It offers users a much more flexible interface
by KrugerDunnings 4y ago
Maybe not interfacing directly with a LLM ala ChatGPT but as an extension of vector search using word embeddings. It offers users a much more flexible interface to explore different combinations of dimensions vs a rigid facet UI interface. Well you can argue that the engineering has shifted to making the model, and that is true, but by keeping it fuzzy there is more room for creative prompt engineering. One challenge will be teaching the user what one doesn't know.
- PaulHoule 4y agoI'd scratch out "word" and replace that with "text" or "sentence" or "document". I saw word embeddings in the mid 2010's as a step backwards and I walked away from a lucrative consulting contract because I couldn't convince the sponsor that it was a total waste to apply word2vec to clinical documents because a significant amount of the meaning of the documents was in out-of-distribution words. (Expecting that to succeed is like a football coach thinking he can shoot the quarterback in the head and still win the game.) For me the train was pulling out of the station with FastText, BERT and similar models that used subword tokens because w/ the last generation of systems you really could show they threw out critical information in preprocessing that would make accuracy a matter of luck at best.