NotebookLM – Ethical Dilemmas of Data Interpretation: A Media Scholar’s and Engineer’s Perspective on an AI Tool

The chapter addresses issues related to the accuracy and reliability of content generated by widely available large language models (LLMs). In discussions on working with AI, much attention is devoted to the generation of unreliable information or the broadly understood phenomenon of model hallucinations. This paper examines the tool NotebookLM—a virtual research assistant that enables collaboration based on previously imported files. This approach minimizes the risk of generating hallucinated content; however, with a carefully selected set of input materials, it also enables the generation of biased outputs. Two examples are presented: the first demonstrates NotebookLM’s resistance to hallucinations, while the second illustrates how a virtual research assistant can be manipulated to produce materials tailored to support a predetermined thesis.