Shashank Gupta

I'm a researcher on the Aristo team at the Allen Institute for AI (AI2) in Seattle.

Broadly, I am interested in building general-purpose AI agents that can:
(1) reason about complex problems through natural language rationalization,
(2) faithfully express their uncertainty in their knowledge and reasoning, and
(3) continuously improve through self-reflection and human feedback.

My current research focus is on AI for Scientific Discovery. I am particularly interested in AI for Math (e.g., Automated Theorem Proving, better foundation models for Math), building agents for finding supporting/contrary evidence from literature, and retrieval-augmented and memory architectures for supporting these use cases.

CV | E-Mail | Google Scholar | Semantic Scholar | Github | Twitter

profile photo
Selected Projects

[NEW] Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs
Shashank Gupta, Vaishnavi Shrivastava, Ameet Deshpande, Ashwin Kalyan, Peter Clark, Ashish Sabharwal, Tushar Khot
ICLR 2024
paper | project | code | data | X (Twitter)


[NEW] Self-Refine: Iterative Refinement with Self-Feedback
Aman Madaan, Niket Tandon, [...], Shashank Gupta, [...], Peter Clark
NeurIPS 2023
paper | project | code | poster
Media: media logo


Sparsely Activated Mixture-of-Experts are Robust Multi-Task Learners
Shashank Gupta, Subhabrata Mukherjee, Krishan Subudhi, Eduardo Gonzalez, Damien Jose, Ahmed H. Awadallah, Jianfeng Gao
Arxiv 2022


Knowledge Infused Decoding
Ruibo Liu, Guoqing Zheng, Shashank Gupta, Radhika Gaonkar, Chongyang Gao, Soroush Vosoughi, Milad Shokouhi, Ahmed Hassan Awadallah
ICLR 2022
paper | code


Exploring Low-Cost Transformer Model Compression for Large-Scale Commercial Reply Suggestions
Vaishnavi Shrivastava*, Radhika Gaonkar*, Shashank Gupta*, Abhishek Jha
*equal contribution
Arxiv 2021
paper | blog post


CogCompNLP: Your Swiss Army Knife for NLP
Daniel Khashabi, Mark Sammons, [...], Shashank Gupta, [...], Dan Roth
LREC 2018
paper | code


Web-scale entity annotation using MapReduce
Shashank Gupta, Varun Chandramouli, Soumen Chakrabarti
HiPC 2013
paper | slides

Template credits: Unnat, and Jon