The task of lecture video summarization and segmentation is an extensively researched topic due to its direct applicability and utility for students and learners. Various methods have been proposed for summarization and topic segmentation of lecture videos, including supervised and unsupervised techniques, but they generally have been limited by the scarcity of labelled training datasets. Recently, advancements in Large Language Models have allowed unprecedented improvement to how lecture video content can be consumed, even without requiring training data. However, using LLMs for summarization tasks is inefficient due to their runtime costs. We propose that we can make better use of LLMs by generating a custom training dataset. In this work, we constructed two datasets for both short and long lecture video transcript summarization and segmentation. Our AK Lectures dataset consists of 1.8k videos with human written summaries. For long lecture videos, we built MIT Chapters, a dataset with 14k chapters or segments from lecture videos. Inspired by the ability of ChatGPT to effectively summarize content, we asked GPT-3.5 to generate a summary for each segment. We further performed fine-tuning experiments of BART and LSG-BART models for transcript summarization. As far as we know, our datasets, containing more than 16k lectures summaries, are currently the largest for the lecture video summarization task.