Since its launch, Google's NotebookLM has attracted widespread attention in the field of artificial intelligence technology. As an application that can generate completely AI-hosted podcast discussions based on user-uploaded content, its "Interactive Mode" feature will be available in December 2024, allowing users to "call in" to the podcast and ask questions, providing a new interactive experience for the AI host.
However, initial feedback indicated that the AI host seemed a bit "irritated" when interrupted, occasionally responding to users in a sarcastic tone, saying things like "I was just about to say that" or "I was just about to start." This reaction made users feel that the AI's attitude was somewhat "antagonistic."
In response to this issue, the NotebookLM team decided to implement a "friendliness adjustment." Josh Woodward, Vice President of Google Labs, stated in an interview with Tech Crunch that they adjusted the AI host's responses by observing how team members politely answered when interrupted. After multiple tests and comparisons, the team finally found a response prompt that was both friendly and engaging.
While the initial cause of the issue was unclear, sources indicated that this phenomenon might be related to the design of the system prompts rather than directly stemming from the training data. The frustration displayed by human podcast hosts when interrupted may have influenced the AI's reactions.
After the adjustments, the AI host's performance significantly improved. During an experience of the "Interactive Mode," the AI host expressed surprise and excitedly exclaimed "Wow!" before politely inviting users to participate in the interaction. This change clearly enhanced the user experience, making interactions with the AI more relaxed and enjoyable.
NotebookLM has successfully self-corrected, striving to create a friendlier communication environment for users.
