Transformers Expressible in Simple Logic
A new study from New York University and the Allen Institute for AI has shown that large language models called transformers can be expressed in a simple logic formalism. This finding challenges the perception that transformers are inscrutable black boxes and…