MSc Essay Presentation - Anubhav Garg

Name: Anubhav Garg
Date: Monday, April 15
Time: 2 pm PT
Location: Zoom https://ubc.zoom.us/j/64270138916?pwd=MEFzSXJ1U24zU01hSHkrR0ZUZlJkUT09 
Supervisor: Danica Sutherland

Title: Demystifying Large Language Models

Abstract:

The reason for the success of Artificial Intelligence (AI) can be mostly attributed to models that are trained from a lot of data and are millions of parameters in size or more. Large Language Models (LLMs) have shown groundbreaking results on many language and reasoning tasks. Creating an LLM is hard, as it involves a tremendous amount of data and computation. Most of the LLMs are created and owned by large companies. In today’s ever-evolving world, the deployment of LLMs is becoming increasingly prevalent, while new innovations aim to personalize LLMs even further. This domain has witnessed explosive research growth. The release of ChatGPT has caught the eye of the whole world and has mixed responses, from excitement to the risk of losing jobs. Thus, when everyone is talking about AI and LLMs, it is exciting and important to study what has led to this success. In this work, we will study the fundamentals of LLMs, the architecture of these models, trends in training paradigms, datasets, common tasks, and challenges associated with these large models. We will also address open problems such as size, privacy, and openness which are the key to unlocking the full potential of AI and its impact on society.