videos

I joined NotebookLM as a podcast guest!

A demonstration of Google's NotebookLM interactive podcast feature, where I participate as a guest in an AI-generated podcast discussion about the Coconut research paper on latent reasoning in large language models. NotebookLM now features an interactive podcast mode (beta) that allows users to join AI-generated podcast discussions about uploaded documents. The Coconut paper introduces a breakthrough method where AI can reason in continuous latent space rather than through explicit chain-of-thought language, potentially improving reasoning for complex problems. I raise philosophical questions about verification of hidden reasoning, and critical concerns about transparency, compliance, and explainability. Latent reasoning creates a black box problem: we see inputs and outputs but cannot observe the reasoning process. I also test the feature's limits by requesting humor, making Monty Python references, and asking the hosts to switch to Dutch, revealing important insights about inclusivity, turn-taking, and accessibility in AI-generated conversations. Despite current limitations where AI hosts can become condescending or defensive, the feature represents a significant step toward more natural, bidirectional conversations with documents and AI systems.