The goal of this project was to automatically create presentation slides using openAI's GPT-3 model.
The first Version uses a Text as input and using GPT-3 summarizes it to bulletpoints. Then, using the bulletpoints, a different GPT-3 Model tries to get the topic from the text. This topic is used to download a matching image Finally the bulletpoints and the image are used to automatically create a PowerPoint file (.pptx)
The second version uses a wikipedia api to get the text of a given topic and generates bulletpoint from it. The picture downloading algortithm is used again and expanded to get multiple matching images Now a presentation with multiple slides corresponding to the subtopics of the wikipedia article is generated
In the third version we wanted GPT to generate the text on its own. We we trained a model to generate bulletpoint from a single topic. This is used in combination with the image downloader to create a PowerPoint presentation file
In the fourth version we wanted GPT to generate the code for its presentation. We therefore switched to a html Presentation using markdown and the jupyter notebook framework for design First a trained GPT model uses the topic input to generate subtopics Then a second model generates text from all the subtopics A third GPT model uses the generated text as input to generate the html markdown code for the presentation At the end a html presentation with one slide per subtopic is automatically generated