Join us for the first AIES event of the 2025-26 academic year! We will introduce AI Ethics & Society, hold a discussion on fine-tuning Large Language Models (LLMs) and engage in a creative exercise with two LLMs.
Since OpenAI's global release of ChatGPT, LLMs have been at the forefront of human interactions with Artificial Intelligence (AI) systems. Trained on massive amounts of data, LLMs assist millions of users worldwide with the most disparate tasks, from productivity to mental health support. In this landscape, the practice of fine-tuning has gained increasing prominence and usage. Also called "adaptation", fine-tuning consists of partially re-training a small subsection of a LLM with additional data on top of its initial pre-training. In this way, LLMs can be adapted to performing specific tasks, such as text summarisation, or to be especially sensitive to a certain kind of perspective. Fine-tuning is substantially cheaper and faster than pre-training, an aspect that certainly contributes to its popularity. But what are the ethical risks, downsides, dangers and shortcomings of fine-tuning - and does it really work?
We will be leading a critical discussion on the topic and then present a LLM that has been fine-tuned on Edinburgh-based perspectives. In a hands-on, creative exercise, participants will be invited to interact with it as well as with its "base" version before fine-tuning.
After 17:00, we will continue the discussion over a pint at Holyrood 9A!
