The University of Michigan recently hosted a collaborative workshop on generative artificial intelligence, organized by the Michigan Institute for Data Science and the Michigan AI Lab. Held at the Weiser Hall, the event attracted participants from both academia and the […]
The University of Michigan recently hosted a collaborative workshop on generative artificial intelligence, organized by the Michigan Institute for Data Science and the Michigan AI Lab. Held at the Weiser Hall, the event attracted participants from both academia and the public.
The workshop kicked off with an introductory address by Bet Ubersejder, the manager of the MIDAS research team. This session was part of a series of collaborative workshops initiated by MIDAS and the Michigan AI Lab. Previous workshops have covered topics such as artificial generative writing and images. The fourth session, held on this occasion, focused on the utilization of large language models, specifically algorithms that generate content based on vast data sets.
Shayne Storks, a doctoral researcher in computer science and engineering at the University, addressed the entire audience at the beginning of the event and assisted in guiding the tutorial throughout the session. Storks, who is currently pursuing a Ph.D., provided insights into various applications of large language models in artificial intelligence.
“These models allow us to generate language based on input data or evaluate the probability of specific text,” Storks explained.
Following his address, Storks led an interactive presentation where participants had the opportunity to test a type of generative artificial intelligence step by step. The presentation was conducted using Google Colab, a Python programming language platform owned by Google. It covered the fine-tuning and execution of large language models. Storks emphasized the importance of constant verification and training of these models due to potential inaccuracies that may arise.
“We cannot blindly trust the outputs of these models in terms of factual accuracy,” Storks cautioned. “We must perform additional verification.”
Te Bulun Altangadas, a bio-statistics student at Rackham University, expressed to The Michigan Daily that the University provides an excellent environment for engaging in artificial intelligence research, especially following the launch of their own generative AI model – U-M GPT – and other internal AI tools.
“I am truly impressed that (the University) has U-M Maizey, which allows for personalization of the GPT model,” Altangadas shared.
In an interview with The Daily, the executive director of MIDAS, Jing Liu, invited all interested individuals to join future workshops in this series and become involved in the MIDAS Student Organizational Council.
“Members of the Student Organizational Council come together to create tutorials and collaborate on projects,” Liu said. “MIDAS connects them to coordinate and even collaborate. Additionally, MIDAS is starting to assist student clubs in connecting with professors and research projects in the industry, providing them with more practical experience.”