Coordinating Multi-Character Conversations between Large Language Models

This project will involve research and development of an LLM framework powered by OpenAI's latest GPT-4 model, for managing natural language conversations between multiple humans and AI entities in group scenarios. This LLM framework is required it identify which characters are the most suitable responders for any provided prompt, and instructing characters when it is their turn to speak. This LLM framework is also responsible for providing important conversation summaries, reducing the risk of catastrophic forgetting or character derailments.

The outputs of this work will be utilised as part of a larger project at Flinders focussed on developing an immersive and interactive “escape room” style experience, intended to be held at Flinders' The Void as part of the Adelaide Fringe Festival in 2025. All students who conduct research to support this demonstration, including the successful recipient of this scholarship will be included as co-authors and/or acknowledgements in a future scientific journal publication.

This project will provide the following student learning experiences:

• Research into the capabilities of Large Multimodal Models (LMMs) capable of interpreting both text and image inputs.
• Development skills working with the OpenAI API, including utilisation of GPT-4 models for text generation and image analysis (GPT-vision), as well as Text-to-Speech and Speech-to-Text with the Whisper model.
• Understanding a variety of modern prompt engineering techniques, including Chain-of-Thought, Self-Refine, Few-Shot prompting, and Automatic-Prompt-Engineering.

This scholarship is valued at $5,000 for 8-12 weeks between December 2024 and February 2025

Applicants must be:

• A third year, honours or masters student at Flinders University during 2024
• Located in South Australia for the majority of the scholarship period, and able to attend regular in-person meetings at Flinders University campus
• Intermediate knowledge of Python programming
• Strong enthusiasm to learn and develop programs that utilise LLM technology (prior experience working with OpenAI's API is desirable)