DiveDeepAI trained GPT-3 to generate human-like text. GPT-3 has been used to generate articles, stories, dialogue etc from a small amount of input text, allowing it to generate large amounts of copy with an exceptional quality.
A stealth educational content providers who needed summarized video transcripts of their courses.
The team had to extract the correct transcription of videos with was a very major challenge. Moreover, moving forward we had to find the optimal sentiment analysis that required a lot of time and effort. Last but not the least was to design database solutions for the summaries extracted from transcription of videos. A special database format was designed to store these summaries.
It is an efficient system that provides one liner summaries of different chunks of a video that are separated bases on the context like keywords, sentiments etc. It starts working by getting correct transcription of the complete video. Once, we get the transcription its separated into parts based on keywords and sentiments. The paragraph obtained is summarized into one line and then finally saved into the database in a specified format. Furthermore, we also use this process to get the sentiment analysis from different parts of the video.
The impact of this project is to accelerate the browsing of a large collection of video data while also achieving efficient access and representation of the video content. Users can make quick decisions about the usefulness of the video by watching the summary.