Teaching LLMs the Value of Cooperation

Tanmoy Chakraborty, Associate Professor of Electrical Engineering and Associate
Faculty Member, Yardi School of AI, Indian Institute of Technology Delhi.

– 2 February 2024

Talk summary: Large language models, despite their astounding reasoning abilities, are not faithful problem solvers. While their abilities are strongly correlated with scale, even humongous models like GPT-3.5 or GPT-4 can become inconsistent reasoners. Recent advances in verbose prompting techniques like chain-of-thought try to elicit step-by-step decomposition so that the model can solve a sequence of simpler problems to finally reach the goal. Augmenting external tools like web search or calculators has also been proposed to offload deterministic tasks. However, foundational language models learn neither problem decomposition nor tool-usage.

In this talk, Tanmoy Chakraborty presented potent solutions towards offloading reasoning subtasks in the case of mathematical problem solving: how does one teach an auxiliary (and potentially frugal) language model to coordinate with black-box solvers, symbolic or language model-based, to successfully answer mathematical problems? This talk focussed on successfully teaching language models to perform reasoning from non-human feedback and how rewards beyond just the correctness of the final answer are essential for better learning.

Speaker bio:  Tanmoy Chakraborty is an Associate Professor of Electrical Engineering and an Associate Faculty Member of the Yardi School of AI at the Indian Institute of Technology (IIT) Delhi. He leads the Laboratory for Computational Social Systems (LCS2), a research group specialising in natural language processing and computational social science. His current research primarily focuses on empowering frugal language models for improved reasoning, grounding, and prompting and applying them specifically to two applications — mental health counselling and cyber-informatics. Tanmoy obtained his PhD in 2015 from IIT Kharagpur as a Google PhD scholar. Subsequently, he worked as a postdoctoral researcher at the University of Maryland, College Park and as a faculty member at the Indraprastha Institute of Information Technology Delhi. Tanmoy has received numerous awards, including the Ramanujan Fellowship, PAKDD Early Career Award, ACL’23 Outstanding Paper Award, IJCAI’23 AI for Good Award, and several faculty awards/gifts from many industries like Facebook, Google, LinkedIn, JP Morgan, and Adobe. He has authored a textbook on Social Network Analysis.

[Talk organised in collaboration with the Department of Computer Science and Automation]