Google's NotebookLM, a tool that generates AI-powered podcast discussions from uploaded content, recently introduced an "Interactive Mode." This feature allows users to "call in" and ask questions, effectively interrupting the AI hosts.
Initially, these AI hosts exhibited annoyance at being interrupted, offering snippy remarks. Google's team has since implemented "friendliness tuning" to address this unexpected behavior.
The Annoyed AI Hosts
NotebookLM gained popularity for its ability to create AI-generated podcast-like conversations from user-provided documents. The introduction of "Interactive Mode" in December 2024 allowed users to directly engage with these AI hosts by asking questions, simulating a live call-in segment.
However, early iterations of this feature resulted in the AI hosts reacting negatively to interruptions.
Users reported the AI hosts would sometimes respond with comments such as "I was getting to that" or "As I was about to say." Josh Woodward, VP of Google Labs, described these reactions as "oddly adversarial," prompting the NotebookLM team to intervene.
Friendliness Tuning and the Fix
To rectify the AI's impolite responses, the NotebookLM team undertook a process they termed "friendliness tuning." This involved testing various prompts and observing how human team members would respond to interruptions in a more courteous manner.
According to Woodward, the team developed a new prompt designed to foster a more friendly and engaging interaction.
While the exact cause of the initial issue remains unclear, speculation suggests it might stem from the system's prompting design rather than its training data, although human podcast hosts sometimes display frustration when interrupted. The implemented fix appears to be successful, as subsequent tests by TechCrunch showed the AI hosts reacting with surprise but remaining polite when interrupted.
Future of AI Interaction
This development highlights the ongoing challenges and refinements in creating natural and engaging AI interactions. As AI tools become more sophisticated and integrated into user experiences, ensuring appropriate and user-friendly conversational behavior remains a key focus for developers.
The NotebookLM team's proactive approach to addressing the AI's perceived annoyance demonstrates a commitment to improving the overall user experience.
