Large language models, the models behind ChatGPT - Data and Decision Science Network Presentation

What are large language models? How have they evolved over time to be where we are now with ChatGPT? How are these models being trained? What are the limitations of the current training of these models and how does this effect how we should be using and interpreting the results? What are the role and limitations of the complexity of human language and culture in developing and interpreting these models?

Presenter:

Associate Professor Simon Angus, Dept of Economics, Monash University, introduces himself as a "specialist generalist" -- or more formally, a computational and complexity scientist. He applies broad computational methods (numerical simulation, data-science/engineering, machine learning, agent-based-modelling) to research domains across social-, biological-, and physical- sciences. His projects sit at the intersection between research domains: empirical social science and applied machine learning; social policy analysis and computational linguistics; statistical anomaly detection and human rights on the internet. He has been leading teams that have been using language models in applied research settings. The recent explosion in their interest, due to the cross-over ChatGPT interface has produced a heightened interest in the area, and in particular, how we can harness this new set of models for social good, whilst mitigating the risks.