A test of Google's NotebookLM software reveals the AI's tendency to hallucinate and expand brief text into lengthy, fabricated discussions. After being fed a 21-second neutral prompt about Donald Trump and Kamala Harris under the pseudonym "Seymour Butts," the AI generated a five-minute podcast featuring "deep dive" analysis that was never in the source material. The experiment demonstrates the current limitations and biases inherent in automated AI summarization tools.
