On September 17, AI Sweden in collaboration with RISE, Research Institutes of Sweden, and the National Library of Sweden, arranged a webinar about GPT-3.
The two hour long event was splitted in two parts. During the first hour, Magnus Sahlgren, Head of Natural Language Processing at RISE, Love Börjeson, Director of KBLab, and Martin Malmsten, Senior Data Scientist, KBLab at National Library of Sweden, gave an introduction to GPT-3. Questions answered included what a language model is in the first place, what's special with GPT-3 and how it can be used. More technical aspects of GPT-3's inner working was then presented by Ariel Ekgren, Research Scientist at RISE.
The webinar was moderated by Johanna Bergman and Anders Thoresson, AI Sweden.
Content
Introduction to GPT-3
- What is a language model and what distinguishes GPT-3 from other models?
- How can it be applied, and how can the Swedish AI ecosystem benefit from it?
- What challenges does the technology present from an ethical, democratic and sustainability perspective?
Welcome and introduction, Johanna Bergman, AI Sweden
Magnus Sahlgren, Head of Natural Language Processing at RISE
Love Börjeson, Director of KBLab, National Library of Sweden
Martin Malmsten, Senior Data Scientist, KBLab at National Library of Sweden
Technical Session
A technical deep dive including how GPT-3 works, when it performs well and when it fails, and how it is possible to train such a large language model.
Q&A Technical session