From Slow to Fast Thinking for LLMs

The partner, Boson AI Inc., is a leading AI solutions provider, specializing in customized model serving for businesses. Through this project, Boson AI aims to tackle key challenges in improving the reasoning capabilities of large language models (LLMs).
While techniques like Chain-of-Thought prompting have shown promise in improving LLM performance on complex tasks, they remain limited in scope and introduce significant computational and latency overhead. This project seeks to improve reasoning efficiency by leveraging recent advances in LLM fine-tuning and alignment, enabling faster, more efficient inference without compromising reasoning accuracy.
The anticipated benefits for Boson AI include significant cost savings, higher customer satisfaction, and enhanced regulatory compliance. Ultimately, the project’s success is expected to strengthen Boson AI’s position as a leader in scalable and efficient AI systems, while advancing its broader research strategy in next-generation generative models.

Faculty Supervisor:

Xujie Si

Student:

Partner:

Boson AI

Discipline:

Computer science

Sector:

Education; Information and cultural industries

University:

University of Toronto

Program: