Common Sense Reasoning in NLP with Vered Shwartz - #461
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
00:00
Unpacking Language Model Biases
This chapter delves into the reporting biases in language models that arise from different training corpora, such as Wikipedia and web data. It highlights the challenges of prompting these models, the variability in responses, and explores techniques for improving prompt effectiveness. The discussion also connects the complexities of plausibility in model outputs to cognitive science literature, emphasizing the need for better methodologies to assess statement validity.
Transcript
Play full episode
