WA Friday Lunch Talks are monthly meetings with presentations of current research results or research in progress by WA faculty, staff, or PhD students. We welcome all to a talk and debate: "Opportunities & risks of large language models for university education" by mgr Kacper Łodzikowski and mgr Mateusz Jekiel (Friday, 3 March 2023, 13:15-14:45).
Abstract
People use Artificial Intelligence (AI) for languages on a daily basis through search engines, customer service chatbots, machine translation, or automated captioning of videos. In the past 5 years, some of those products started relying on Large Language Models (LLMs), i.e. systems that learn how to understand natural language. LLMs were so tightly integrated into those products that the general public was not aware of their potential until the release of ChatGPT, a writing assistant that completes a wide range of tasks via natural language generation.
We will start this session with a talk describing how LLMs work, focusing on a particular model called GPT (Generative Pre-trained Transformer). We will demonstrate how the model’s semantic understanding ability relies on the relative mapping of words in a latent space. We will describe how GPT-3 was combined with another AI technique called reinforcement learning from human feedback to create ChatGPT. We will conclude by illustrating the strengths and limitations of (Chat)GPT for real-life use cases.
The talk will be followed by a debate on the opportunities & risks of LLM-based tools such as ChatGPT for university education. Is it ethical to use ChatGPT for completing homework, incl. writing essays? Should students disclose the use of the tool? Should teachers allow it? Does computer literacy in the 2020s now include the ability to use AI writing assistants? We invite all staff and students to share their thoughts.
About the speakers
|
|
