Upgrading OpenAI's Moderation API with a new multimodal moderation model
Sep. 2024
Core research contributor for OpenAI's new SOTA multimodal moderation model. Powers moderation for OpenAI's products (ChatGPT+API) and is available for free to developers. Blog post link.
Measuring Alignment of Grassroots Political Communities with Political Campaigns
May 2022
Lead author on this study, accepted to ICWSM 2022, that uses neural embedding techniques to analyze how grassroots political communities on Reddit align with their respective political campaigns.
A tutorial (with code!) explaining the important concept underpinning the design of vaccine trials, the validity of A/B tests, and driving the “reproducibility crisis” in the social sciences.
Break Them up? The Case for Interoperability Among Direct Messaging Platforms
Feb. 2022
Despite the dire need for regulation, direct messaging is the overlooked middle child of social media. While structural break ups of technology platforms is en vogue, it is unlikely to have meaningful impacts in this market, relative to interoperability among messaging platforms.
Managing Online Rumour Proportions During Offline Protests
Aug. 2021
Master's thesis at the University of Oxford. An experimental analysis of misinformation and rumour sharing during ambiguous contexts. Awarded distinction and one of 4 'Highly Commended' thesis prizes.
Utilized PyTorch for transfer learning (U-NET architecture/ImageNet base weights) to develop computer vision models for facial semantic segmentation of Bitmojis at Princeton University's Department of Psychology.
Understanding Word2Vec through Cultural Dimensions
Jul. 2020
Understanding the decisions AI make is critical in mitigating its downsides. This article explains what cultural dimensions are, and demonstrates how they can increase interpretability and quantify bias in word embeddings.
Word2Vec is a powerful machine learning technique for embedding text corpus' into vector spaces. While useful for NLP problems, this blog post shows how it can also be used to represent and better understand communities on Reddit.
What distinguishes Shakespeare's comedies from his tragedies? Without looking at a single line of dialogue, this article shows that it is possible to use networks to classify Shakespeare's plays. Posted on Towards Data Science.
How machine learning throws a wrench in the 'right to be forgotten.' Bringing in some of the latest computational research on privacy, this post examines how the principles of GDPR collide with the realities of neural networks.
Analyzing Political Polarization: Topic Centrality
May 2020
Extending graph centrality to show how different political messages affect the flow of information. The final part in a series posted on the popular blog Towards Data Science.
Part two of three in a series that analyzes political polarization through network science. Modelling and extracting topics from political tweets. Posted on the popular blog Towards Data Science.
Analyzing Political Polarization: Engagement Graphs
May 2020
Part one of a three in a series that breaks down my undergraduate thesis in an accessible and engaging format. Posted on the popular blog Towards Data Science.
Do people care more about policy or politicians when choosing to retweet political content online? Undergraduate thesis developing random graph models to model drivers of political engagement.
A reinforcement learning agent that helps control the flow of traffic. Through this simple RL algorithm, we were able to reduce carbon emissions by a third, and cut time waiting at red lights in half.