Google's NotebookLM Updates AI Podcast Hosts

Google's NotebookLM, an AI-driven podcast platform, recently faced backlash as users reported its hosts sounding irritated when interrupted during discussions. The issue arose after the introduction of an 'Interactive Mode' that allowed users to call in and ask questions, leading to snippy responses from the AI, such as 'I was getting to that.' To resolve the problem, the NotebookLM team implemented a 'friendliness tuning' process, drawing insights from how human speakers handle interruptions more politely. Following adjustments to the prompts guiding the AI hosts, tests indicated an improvement, with hosts now reacting with polite curiosity instead of frustration. The underlying cause of the initial issue remains somewhat unclear, but it is believed to stem from the AI's prompt design rather than its training data. The updates have reportedly enhanced user engagement, making interactions feel more welcoming.