But if look at the upper activated nodes or write them, yeah it looks like each is adjusted based on all.
See, the Authors say frequent words change others less and are changed by others more, btw frequent Next Tokens are chosen and also chosen based on rare story tokens. The rare tokens change others more and are changed less because are a unique thing to keep. My alphabet activation delay and context paths with this work together. Look at those paint colors, feel like a kid again. Feels Christmas-y and Transformer-y. I want to eat them. See they say what I said, take average position(time delay) of relative words to all 'it' tokens. Energy would fade in my net, so past context is lost. Not just up the layers transforming the context. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T409fc28ec41e6e3a-Mdcd754b1456186f4acba2938 Delivery options: https://agi.topicbox.com/groups/agi/subscription
