Just read this paper "Grokked Transformers as Implicit Reasoners". The authors really went all-in on understanding how transformers learn to reason implicitly through this "grokking" phenomenon. They focused on two key reasoning tasks: composition (like chaining facts) and… https://t.co/lYKBqsGn46 https://t.co/CsKK8g6sUh
— Avi Dhaliwal (@champydaku) Jul 5, 2024
from Twitter https://twitter.com/champydaku
July 05, 2024 at 07:35PM
via IFTTT
No comments:
Post a Comment