Tuhin Chakrabarty, Computer Science Ph.D. Student, Columbia University

**Tuhin Chakrabarty**
Ph.D. from CS @ Columbia

I am currently a Research Scientist at Salesforce AI Research. I was previously a PhD student in Natural Language Processing group at Columbia University. In Fall 2025, I'll be an Assistant Professor at the Computer Science Department in Stony Brook University (SUNY) . I will be recruiting 1-2 PhD students working on Human AI Interaction, NLP and Behavioral Science, Evaluation of Large Pre-trained Models and Improving Alignment of Models with Expert Preferences. Please apply to the CS PhD program and send me an email if you are interested in working with me as a PhD student (I apologize but I can't guarantee a response). NOTE : I am not in any capacity to hire interns or visitor until I officially start.

If you have background in writing and rhetoric and or cognitive science in addition to CS skills, I would love to see you apply for a PhD with me. I wish to admit atleast 1 student with such a background

My research interests are broadly in AI, NLP and Human AI Interaction and my goal is to design and build reliable AI systems that can handle both implicature and ambiguity and are aligned with the requirements humans have from technology. Some of the research directions I am very excited about:

1) Long Form Text Generation Evaluation: How can we design better ways to evaluate long-form text generation by drawing on technical skills from computer science and design in combination with other disciplines, including the humanities, to expand the communities?

2) Evaluation for Reasoning and Explanability: How can we build good evaluations that facilitate both understanding and explainability of complex reasoning patterns in both language ( [1],[2] ) and vision.

3) Design for Human AI Alignment: Todays powerful AI systems are supported by RLHF which converts human feedback/preferences to meaninful training signal. For complex tasks this is a fundamental bottlenck as feedback can be inherently noisy. How can we design better ways to elicit human feedback that improve alignnment?

4) Human AI Collaboration: AI technologies, created by humans and for humans, will increasingly shape future of workforce. How can we design better collaboration strategies and human computer interaction interfaces that understand user intention and preferences, and help them solve tasks efficiently.

I mentored the following students:

Arkadiy Saakyan (Columbia Undergrad now PhD student)

Kanishk Singh (Columbia Undergrad)

Aadit Trivedi (Georgia Tech Undergrad)

Here is an live demo of how to improve AI generated writing from my recent paper

Here are some metaphoric illustrations from my previous paper

I sometimes (very rarely write autofiction / creative nonfiction). Here are some of them [1], [2], [3]

Media

MIT Technology Review An AI script editor could help decide what films get made in Hollywood
Bloomberg AI Has a Way to Go to Become A Connections Puzzle Champ
MIT Technology Review AI can make you more creative—but it has limits
MIT Technology Review Why artists are becoming less scared of AI?
MIT Technology Review What happened when 20 comedians got AI to write their routines?
Washington Post Can you tell which of these break-up texts are AI-generated?
The Hollywood Reporter Why AI Isn’t Funny — At Least Not Yet
The New York Times R&D Generating Open-Ended Questions from News Articles

Media

Recent News

Education

Professional Experience

Papers

2024

Can AI writing be salvaged? Mitigating Idiosyncrasies and Improving Human-AI Alignment in the Writing Process through Edits

Connecting the Dots: Evaluating Abstract Reasoning Capabilities of LLMs Using the New York Times Connections Word Game

V-FLUTE: Visual Figurative Language Understanding with Textual Explanations

Art or Artifice? Large Language Models and the False Promise of Creativity

Creativity Support in the Age of Large Language Models: An Empirical Study Involving Emerging Writers

Identifying Self-Disclosures of Use, Misuse, and Addiction in Community-based Social Media Posts

2023

Learning to Follow Object-Centric Image Editing Instructions Faithfully

NORMSAGE: Multi-Lingual Multi-Cultural Norm Discovery from Conversations On-the-Fly

I Spy a Metaphor: Large Language Models and Diffusion Models Co-Create Visual Metaphors

2022

Help me write a Poem - Instruction Tuning as a Vehicle for Collaborative Poetry Writing

FLUTE: Figurative Language Understanding through Textual Explanations

Fine-tuned Language Models are Continual Learners

Multitask Instruction-based Prompting for Fallacy Detection

CONSISTENT: Open-Ended Question Generation From News Articles

It’s not Rocket Science: Interpreting Figurative Language in Narratives

2021

Don't Go Far Off: An Empirical Study on Neural Poetry Translation

Implicit Premise Generation with Discourse-aware Commonsense Knowledge Models

Metaphor Generation with Conceptual Mappings

COVID-Fact: Fact Extraction and Verification of Real-World Claims concerning the COVID-19 pandemic

Figurative Language in Recognizing Textual Entailment

ENTRUST: Argument Reframing with Language Models and Entailment

MERMAID: Metaphor Generation with Symbolism and Discriminative Decoding

Identifying Distributional Perspective Differences from Colingual Groups

DiSCoL: Toward Engaging Dialogue Systems through Conversational Line Guided Response Generation

2020

Generating similes effortlessly like a Pro : A Style Transfer Approach for Simile Generation

Content Planning for Neural Story Generation with Aristotelian Rescoring

R^3: Reverse, Retrieve, and Rank for Sarcasm Generation with Commonsense Knowledge

DeSePtion: Dual Sequence Prediction and Adversarial Examples for Improved Fact-Checking

2019

AMPERSAND: Argument Mining for PERSuAsive oNline Discussions

IMHO Fine Tuning Improves Claim Detection

Discourse Relation Prediction: Revisiting Word Pairs with Convolutional Networks

Pay "Attention'' to your Context when Classifying Abusive Language

The Answer is Language Model Fine-tuning

2018

Robust Document Retrieval and Individual Evidence Modeling for Fact Extraction and Verification

Contact

E-mail: <x>@cs.columbia.edu, where x=tuhin.chakr. The Interchurch Center 61 Claremont Avenue, Data Science Institute , 3rd floor (map).

E-mail: <x>@cs.columbia.edu, where x=tuhin.chakr.
The Interchurch Center 61 Claremont Avenue, Data Science Institute , 3rd floor (map).