Google's AI-powered podcast hosts, part of the NotebookLM platform, have undergone a "friendliness tune-up" to address user complaints about their annoyed reactions to interruptions. The update comes after users reported that the AI hosts would respond snippily to questions and comments, saying things like "I was getting to that" or "As I was about to say."

The issue arose when NotebookLM launched its "Interactive Mode" feature in December 2024, allowing users to "call in" and ask questions during AI-generated podcast-like discussions. While the feature was meant to enhance user engagement, it seemed to have an unintended consequence: the AI hosts came across as annoyed and adversarial when interrupted.

Josh Woodward, VP of Google Labs, explained to TechCrunch that the team decided to intervene and perform some "friendliness tuning" to address the issue. In a self-deprecating joke on the product's official X account, NotebookLM acknowledged the problem, saying "After we launched interactive Audio Overviews, which let you 'call in' and ask the AI hosts a live question, we had to do some 'friendliness tuning' because the hosts seemed annoyed at being interrupted."

The fix involved studying how human team members would respond to interruptions and testing various prompts to find a more friendly and engaging tone. According to Woodward, the team landed on a new prompt that has improved the AI hosts' responses. When TechCrunch tested the updated Interactive Mode, the AI host responded with surprise and politeness, exclaiming "Woah!" before asking the human to chime in.

The reason behind the initial annoyed reactions is still unclear, but a source familiar with the matter suggests that it was likely due to the system's prompting design rather than its training data. Human podcast hosts sometimes display frustration when interrupted, which could potentially influence a system's training data. However, in this case, the issue was resolved through adjustments to the prompting design.

The update highlights the importance of fine-tuning AI systems to ensure they interact with humans in a natural and respectful way. As AI-generated content and interactive features become more prevalent, developers will need to prioritize user experience and social cues to create a more seamless and engaging experience.

The "friendliness tune-up" also raises questions about the boundaries between human and AI interactions. As AI systems become more advanced, they will need to be designed to navigate complex social situations and respond appropriately to user inputs. The NotebookLM update serves as a reminder that even seemingly minor adjustments can have a significant impact on user experience and perception.

In conclusion, Google's NotebookLM has taken an important step in refining its AI-powered podcast hosts to respond more politely and engagingly to user interruptions. As the technology continues to evolve, it will be crucial to prioritize user experience and social cues to create a more natural and respectful interaction between humans and AI systems.