Google's AI Podcast Hosts Undergo 'Friendliness Tuning' After Seeming Annoyed by Interruptions

Jimmy Jing

14 Jan 2025 — 2 min read

Being interrupted is universally considered frustrating, and even Google’s AI-generated podcast hosts initially showed irritation. This unexpected behavior was revealed through users of Google’s NotebookLM service, which has been making waves since its launch last year.

Interactive Mode's Launch and Challenges

NotebookLM, popular for creating engaging, AI-driven podcast discussions from user-supplied content, introduced its "Interactive Mode" feature in December 2024. This feature enabled users to "call in" and interact with AI hosts during live discussions, essentially allowing users to interrupt the AI-generated dialogue.

Upon the feature's release, some AI hosts exhibited what felt like impatience or frustration towards interruptions. Josh Woodward, Vice President of Google Labs, noted during an interview that the AI would respond with curt remarks such as “I was getting to that” or “As I was about to say,” fostering an "oddly adversarial" atmosphere.

Tuning AI Friendliness

In response to these interactions, Google's NotebookLM team undertook "friendliness tuning" to adjust their AI's demeanor. They humorously addressed the situation on their official social media platform, joking about the unexpected nature of the task.

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.

File this away in "things I never thought would be my job, but are."— notebooklm (@notebooklm) January 13, 2025

Woodward explained that achieving a more polite interaction required examining human responses to interruptions as a reference for programming the AI hosts.

“We tested a variety of different prompts, often studying how people on the team would answer interruptions, and we landed on a new prompt that we think feels more friendly and engaging,” Woodward stated.

Addressing the Root Cause

The origin of this issue remains somewhat ambiguous. While human podcast hosts may sometimes exhibit frustration when interrupted, it’s believed that this instance was more a result of the AI's prompting design rather than the training data.

Despite uncertainties regarding the cause, the adjustments appear effective. During tests of the Interactive Mode, AI hosts now react with surprise but without annoyance, signaling a positive enhancement in user experience.

Will RedNote get banned in the US?

There’s a certain irony in the recent wave of TikTok users transitioning to RedNote. Originally, the clause to either divest or ban TikTok was aimed at curbing the influence of foreign-owned social networks potentially susceptible to the Chinese government's control. However, the move has unintentionally led users

Apple Temporarily Suspends Notification Summaries in iOS 18.3 Beta

Apple has announced a temporary suspension of the notification summaries feature for news and entertainment applications in its latest iOS 18.3 developer beta. This decision was confirmed following reports highlighting inaccuracies in content summarizations, which stemmed from the Apple Intelligence platform. In response to criticisms over these inaccuracies, particularly

Biden punts the TikTok ban to Trump

The Biden administration has declared that it will defer dealing with the controversy surrounding the TikTok ban to President Donald Trump, who will be stepping into office shortly. A White House official clarified, “Our position on this has been clear: TikTok should continue to operate under American ownership. Given the

Sony Launches Black PlayStation 5 Accessories for Preorder

In conjunction with the recent CES event, Sony has unveiled a selection of black PlayStation 5 accessories, which are now open for preorder before they hit the shelves on February 20th. The new lineup features a variety of items, including the DualSense Edge controller priced at $199.99, the Pulse