Reproducing Deep Learning Models

Rajiv Shah attempted to reproduce an earthquake-predicting deep learning model. His results exposed some issues with the model. Kyle and Rajiv discuss the original paper and Rajiv's analysis.

Guest

Rajiv Shah: When I was in the military, my nickname was the “why guy.” I never wanted to just execute, I wanted to understand the larger goal and find a better solution. That curiosity pulled me into data science, where I helped create millions in value, earned patents and publications, trained hundreds of data scientists, and worked across industries using everything from interpretable models to deep learning. Over the last several years I’ve focused on early‑stage and high‑growth AI companies, especially customer‑facing roles where product, GTM, and solutions meet. I love working directly with customers, turning their vague problems into clear AI solutions, and helping teams ship innovative, practical AI that actually lands inside the enterprise. I’ve been part of GTM and enablement teams at DataRobot, Snorkel AI, Hugging Face, Snowflake, Contextual AI, and now Openhands with a track record of scaling customer growth and technical adoption. I’m at my best in the messy, zero‑to‑one phase: building trust with technical buyers, explaining complex tools in plain language, and helping teams align on what “value” really means. I’m also a strong external communicator with 100+ talks and keynotes, publications across computer science, law, and communications, and a public platform (@rajistics) focused on practical AI. Recently, I have started to share back my experience with a course on framing AI problems, learn more over at aiframer.dev