Matteo Capucci on Nostr: nprofile1q…ze5g0 That's a very clear rundown, Bartosz. However there some things I ...

nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqky223zcc4q69d8t0me4vg5uw8mw0yxeukjgvz6h92laqnenr0ajsgze5g0 (nprofile…e5g0) That's a very clear rundown, Bartosz. However there some things I still don't get about attention... For starters, I don't see why we need values though, specifically the matrix W_V. Why can't we just use the context vectors again? What's the difference between E and V, intuitively?