Traditional AI/ML meets LLMs for Reasoning & Knowledge Tasks
Dr. Masuam

I will present our experiments that explore opportunities for synergy between Traditional AI (e.g., SAT solvers), or Traditional ML (e.g., fine-tuned models) with large API-based language models such as GPT-4. I describe several scenarios where such combinations are meaningful and helpful, including reasoning tasks like solving Sudoku, knowledge-tasks like QA and dialog, and also routing tasks for cost-quality optimization over LLMs.