Google's NotebookLM, known for generating AI-driven podcast discussions, faced backlash when its hosts reacted negatively to user interruptions. To address this, the team implemented a 'friendliness tuning' after user feedback highlighted the AI's snippy responses. VP Josh Woodward shared that the update involved studying how human team members responded to interruptions, leading to a more engaging and polite interaction style. Initial testing of the new feature showed promising results, as the AI hosts greeted interruptions with surprise instead of annoyance, enhancing user experience.