from Hacker News

Selection-Inference: Exploiting Language Models Interpretable Logical Reasoning

by silencedogood3 on 5/20/22, 4:46 PM with 2 comments

• by trenchgun on 5/21/22, 9:42 AM

  Twitter thread by one of authors: https://twitter.com/mpshanahan/status/1527686538720595969

• by trenchgun on 5/21/22, 9:36 AM

  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?