In a recent study, the University of California Berkeley’s David Bamman and his team sought to uncover the training data used by GPT-4, a chatbot designed for understanding literary texts. The research involved examining the bot’s recollection of passages from hundreds of books in order to determine which books have been memorized in its neural network. The top 50 books found in the bot’s memory constitute an approximation of the chatbot’s canon, which consisted of a significant amount of science fiction and fantasy, as well as popular novels. The study highlights the need for transparency with regard to chatbot training datasets to address intellectual property issues and the potential biases unintentionally instilled in these models. While more work needs to be done to understand the effects of the data used to train chatbots, tapping into diverse literature and expanding the universe of possible narratives represented has the potential to yield more interesting results.
OpenAI is a research group based in San Francisco that is committed to developing computers that can perform cognitive tasks that typically can only be done by humans, such as problem-solving and creativity. The group has developed several artificial intelligence (AI) models, including GPT-3 and GPT-4, which are natural language processing chatbots that can generate human-like text based on prompts.
David Bamman is an information scientist and assistant professor at the University of California Berkeley School of Information. Bamman and his team are focused on investigating the use of algorithms for analyzing literature and cultural trends, with a particular focus on natural language processing. Their research seeks to explore the intersection between language, culture, and technology.
