The author uses GPT-4 as an information extraction model to extract relevant entities and relationships from a video transcript. The transcript is obtained from a nature documentary, specifically about the deep sea ecosystem. The author groups the transcript into sections based on pauses in narration and then uses GPT-4 to extract entities and relationships. The extracted entities are related to biology, chemistry, or archeology, and include animals, locations, and biological entities. The model is able to generalize across different domains and provide consistent results. However, the author notes that the model's performance can vary depending on the input data and that it may hallucinate information such as citations. The extracted relationships are provided in a triple format (Head, Relationship, Tail). The author also uses GPT-4 for entity disambiguation by asking the model to identify which values reference the same entity. The results show that the most mentioned animals are moray eels, lionfish, and brittle stars. The knowledge graph is used as a search engine to retrieve timestamps of sections where relevant entities are mentioned. The author concludes that GPT-4 is a powerful tool for exploring and analyzing different datasets to extract relevant information.