Big AI models used to get better by getting wider or reading longer texts, but those tricks are slowing down.
Selective Steering is a new way to gently nudge a language model’s inner thoughts without breaking its flow or skills.
The paper builds special Turkish legal AI models called Mecellem by teaching them from the ground up and then giving them more law-focused lessons.
Deep Delta Learning (DDL) replaces the usual “add the shortcut” rule in deep networks with a smarter, learnable move that can gently erase old info and write new info along a chosen direction.