Скачать или смотреть [CW Paper-Club] Are Emergent Abilities of Large Language Models a Mirage?

Welcome to CloudWalk's weekly paper-club session, where our R&D team presents interesting research papers.

In this week's session, Danubio Müller presents the paper "Are Emergent Abilities of Large Language Models a Mirage?" published in 2023 by Rylan Schaeffer, Brando Miranda and Sanmi Koyejo from Stanford University.

In this paper, researchers explore the claims of emergent abilities in Large Language Models (LLMs) that are not present in smaller models. They propose an alternative explanation, arguing that the choice of metric during analysis affects the inference of emergent abilities. Through a mathematical model and three analyses, they provide strong evidence that emergent abilities may stem from researchers' analyses rather than being inherent to scaling AI models.

If you're interested in joining our team at CloudWalk, please check out our job openings on LinkedIn! Don't forget to check out the paper, which is available at the link provided below.

Paper Link: https://arxiv.org/pdf/2304.15004.pdf

Recording Date: May 19 2023

00:00 Paper presentation
15:25 Discussion (mid-presentation - questions about emergent abilities)
31:22 Paper presentation (cont.)
54:55 Discussion