How to get AI to sound less drunk: the GPT-3 case study

By Sifted - 2021-02-25

Description

The much-hyped GPT-3 still lacks understanding of the world — but we can anticipate new ways of teaching AI a real understanding of language.

Summary

  • But even with GPT-3, AI is still stuck in Uncanny Valley.
  • It is our ability to apply ‘slow thinking’ that keeps us honest, and is what GPT-3 currently lacks.
  • The Feedback Transformer improves on the original transformer in two ways.
  • Once we achieve that, AI can graduate from a parlour trick into the real business world.

 

Topics

  1. NLP (0.26)
  2. Machine_Learning (0.2)
  3. Backend (0.07)

Similar Articles

Is GPT-3 Islamophobic?. How OpenAI’s Western algorithm

By Medium - 2021-02-03

The scope of this article is to investigate the intersection of Orientalism, as developed by Edward Said, and technology, in the context of OpenAI’s GPT-3 algorithm, which generates coherent text…