University Libraries Present: The Birth and Evolution of LLM Part 2

Event Description

In this workshop, we will continue our historical journey toward today's large language models, such as ChatGPT and Claude. Part 1 showed how far counting can go with Markov chains and N-grams, and where it fails: a counting model cannot handle word sequences it has never seen. In Part 2, we will start with Bengio's Neural Probabilistic Language Model (2003), one of the first neural network language models to outperform N-grams. We will then look at how context changed with RNNs (2010), attention (2014), and the Transformer (2017), the architecture behind today's LLMs. No background in computer science or AI is required, and Part 1 is helpful but not required. The difficult part is the idea that a word can be a list of numbers. Our journey covers roughly 2003 to today.

Location: Melville Library, Central Reading Room, Lab B

Register to attend.

Date Start

Date End