Selection-Inference: Exploiting Language Models Interpretable Logical Reasoning (opens in new tab)

(arxiv.org)

4 pointssilencedogood34y ago2 comments

2 comments

2 comments · 2 top-level

So, they got neural networks to do multistep logical reasoning by adding the result of single step reasoning to the knowledge base?

What is the performance compared to symbolic reasoning?

j / k navigate · click thread line to collapse