NLP is not NLU and GPT-3 - Walid Saba
Published:Nov 4, 2020 19:16
•1 min read
•ML Street Talk Pod
Analysis
This article summarizes a podcast episode featuring Dr. Walid Saba, an expert critical of current deep learning approaches to Natural Language Understanding (NLU). Saba emphasizes the importance of a typed ontology and the missing information problem, criticizing the focus on sample efficiency and generalization. The discussion covers GPT-3, including commentary on its capabilities and limitations, referencing Luciano Floridi's article and Yann LeCun's comments. The episode touches upon various aspects of language, intelligence, and the evaluation of language models.
Key Takeaways
- •Dr. Walid Saba, an NLU expert, critiques deep learning approaches.
- •He emphasizes the importance of a typed ontology and the missing information problem.
- •The discussion covers GPT-3, including its capabilities and limitations.
- •The episode touches upon the evaluation of language models.
Reference
“Saba's critique centers on the lack of a typed ontology and the missing information problem in current NLU approaches.”