- What: A BEAR Special Interest Group (SIG) establishing a cross disciplinary network of academics working on, or with Large Language Models (LLMs).
- Why: To foster knowledge exchange, facilitate the implementation of LLMs on the UoB High Performance Computing (HPC) cluster, and encourage cross disciplinary collaboration
- Where: A Teams channel (email Allison Koh to join) for general introductions, interactions, tips, and questions.
- Who: Anyone either already working with LLMs, or looking to do so in the future.
The study of Large Language Models (LLMs) and their integration into traditional academic research is still very much a wild west. Researchers across disciplines are exploring what the options are, how they are implemented, and how to do this in an academically robust way. At these early stages of discovery, the exchange of experiences and knowledge between researchers can ease this process and even have a significant impact on the quality of the work. Therefore, we are introducing a BEAR LLM Research group to build a cross-disciplinary network of researchers from UoB to foster this exchange, and possibly inspire inter-disciplinary collaborations.
So, why a BEAR SIG? The general consensus at the moment is that the application of LLMs that enable more control and transparency over methods and outcomes is to employ open-source models locally and fine-tune them for a specific application. This is too computationally intensive to do on regular devices, and therefore a high-performance computing cluster like BlueBEAR is essential in enabling such applications. Since this process is not straightforward, a SIG would prevent researchers from having to re-invent the wheel, where others have already done so. Anyone is therefore welcome to join the group, whether they are working on these things now, or planning to do so in the future.
The group will convene on a Microsoft Teams channel, hosted by the Centre for Artificial Intelligence in Government (CAIG). This will act as a largely informal space to get acquainted with other researchers working with these tools. Members are welcome to ask questions and encouraged to share helpful tips, sources, and events. Depending on members' needs and interests, future events can be hosted where projects are presented to one another in more detail. Email Allison Koh at a.w.koh@bham.ac.uk, or message on Teams to be added to the group. Feel free to post a brief introduction of yourself and your specific interests in this field once added.