Two undergraduates have developed Dia, an AI speech model that rivals Google's NotebookLM, despite having minimal prior experience in AI. The model allows users to generate podcast-style clips with customizable voice features, utilizing Google's TPU Research Cloud for training. With 1.6 billion parameters, Dia can produce realistic dialogues and clone voices, but it also raises concerns about misuse for disinformation. Nari Labs, the creator, plans to enhance Dia with a social platform aspect and expand language support. Investors are optimistic about the potential of voice AI technology, which has gained significant funding recently.