Yamil Ricardo Velez

Assistant Professor of Political Science at Columbia University specializing in tailored experiments, public opinion research, and political psychology.

About Me

I am an Assistant Professor of Political Science at Columbia University, specializing in political psychology with a focus on attitude and belief change. My research examines how the quality of political information influences political engagement and representation, paying particular attention to the impact of misinformation within immigrant communities.

My most recent work examines the potential for generative artificial intelligence to improve our understanding of public opinion, leveraging personalized surveys that adaptively respond to voters. I have been developing LLM-powered methodology since 2021, beginning with GPT-2 for placebo generation and expanding into adaptive surveys, tailored experiments, and retrieval-augmented tools. My work has been published in the American Political Science Review, American Journal of Political Science, Journal of Politics, British Journal of Political Science, PNAS, and Political Analysis.

LLM Research Timeline
2021
Automated placebo generation (GPT-2)
2022
Tailored experiments to study attitude polarization
2023
Crowdsourced adaptive surveys, tailored conjoints, Voting Advice Bots
2024
Eliciting personally relevant beliefs using LLMs
2025
Proof-of-life verification for surveys
2026
More coming soon

News & Updates

Discussed applications of large language models to experimental design, including tailored experiments and tools for mapping the latent network of beliefs and attitudes.

Presented at this event hosted by the NYU Abu Dhabi Institute in New York, discussing the intersection of data science, artificial intelligence, and political research.

Gave a talk titled "Generative AI and Respondent-Centered Social Science," discussing applications of Generative AI to survey design and randomized experiments.

Presented Crowdsourced Adaptive Surveys, exploring how large language models and multi-arm bandit algorithms can optimize survey questions in real-time based on participant feedback.

Delivered a presentation to the New York Chapter of the American Association for Public Opinion Research on the innovative use of Generative AI in surveys and experiments. Discussed the potential and challenges of using LLMs for adaptive polling and tailored experimental designs.

Publications & Research

Experiment

Issue-Based Microtargeting

AI-generated ads tailored to voters' issue priorities outperform demographic and personality-based microtargeting in audio-based conjoints.

Experiment

Confronting Core Issues

A critical assessment of attitude polarization using tailored experiments.

Download Paper
Experiment

Tailored Conjoints

Exploring the impact of deeply-held issues among Latinos using personalized conjoint experiments.

Survey Research

Crowdsourced Adaptive Surveys

Real-time survey optimization using multi-arm bandits. Open-source research implementation available.

Survey Research

Retrieval Augmented Representation

Developing dynamic measures of responsiveness through retrieval augmented generation.

Experiment

Placebo Generation using LMs

Using language models to generate placebo text for survey experiments, reducing researcher degrees of freedom.

Experiment

VAA Bot: Chatbot-Driven Voting Aid

Using LLMs and retrieval-augmented generation to create personalized voting aid applications.

Download Paper

Tools & Tutorials

Tutorial

Tailored Experiments Generator

A web application for generating tailored experimental designs in Qualtrics.

Generate Tailored Experiment
Tutorial

AI-Powered Follow-Up Questions Tutorial

A tutorial on how to use AI to ask follow-up questions in Qualtrics.

See Tutorial
Tool

Pulse: Proof of Life

A Qualtrics plugin that verifies physical human presence to combat AI survey fraud.

View Tool & Documentation

Get In Touch

Have questions or want to collaborate?