Abstract: Large generative language models are transforming entire industries due to their wide range of use cases, particularly because they enable entirely new forms of user interaction. The focus of this project is to enhance the emerging capabilities of these models in logical reasoning and planning (fine-tuning, reinforcement learning, in-context learning). Complex tasks are to be automatically broken down into smaller subtasks, which are then solved through the automated use of specialized tools (libraries, e.g., for arithmetic, specialized search engines, databases).
Our practical goal is the automated handling of complex research tasks and the (partial) automation of routine activities in order to free up employee resources for tasks of higher importance and urgency. The project evaluates applications in tax consulting and accounting as well as auditing.
Project Partners: UTN, Intrafind, TH Deggendorf