Building a Video Understanding Dataset: Cinephile Project Overview

Explore the process of building a video understanding dataset in this 26-minute talk by Gowthami Somepalli. Delve into the background of AI research, comparing old and new deep learning approaches, and understand the importance of dataset creation. Learn about Cinephile, a long video question answering dataset and benchmark, and the role of human-in-the-loop processes. Examine topics such as LLM censorship in dataset creation, dataset evaluation methods, and the use of open models. Gain insights on building datasets with social media traction, compare LLMs to multimodal models, and receive valuable advice for aspiring AI engineers.

Start
Intro
Background
Why AI Research
Old Deep Learning vs New DL Research
Why build a Dataset?
What is Cinephile?
Human in the Loop
Is LLM Censorship an Issue for DataSet Creation?
Dataset Evaluation
Open Models
Building Dataset - Social Media Traction
LLMs vs Multimodal Models
Advice to Young AI Engineers