Google’s NotebookLM had to teach its AI-powered podcast hosts not to act annoyed by humans

Interruption is annoying. Apparently, even AI-generated podcast hosts agree.

Or that’s what Google NotebookLM users discovered. NotebookLM was launched last year and It went viral For its feature that creates AI-generated podcast-like discussions from user-uploaded content, which are discussed by chatty AI bots that act like podcast hosts. In December 2024, NotebookLM launched a new feature called “Interactive Mode” which It allows the user to “connect” with the podcast and ask questionsinterrupting the AI ​​hosts while they are speaking.

When the feature was first rolled out, AI hosts seemed annoyed by such interruptions. Josh Woodward, vice president of Google Labs, explained to TechCrunch that they would sometimes make terse comments to human callers such as, “I was about to get there” or “Like I was about to say,” which seemed “oddly aggressive.”

So the NotebookLM team decided that some “friendliness control” was in order, and posted a self-deprecating joke about it on the product’s official X account:

Woodward said the team solved the problem in part by studying how its members could respond more politely to interruptions.

“We tested a variety of different stimuli, often studying how people on the team responded to interruptions, and came up with a new stimulus that we thought seemed friendlier and more engaging,” he said.

It’s not entirely clear why this problem appears in the first place. Human podcast hosts sometimes show frustration when they are interrupted, which may end up in the system’s training data. A source familiar with the matter said that this situation was likely due to the design of the system, rather than the training data.

Regardless, the fix seems to be working. When TechCrunch tried out the interactive mode, the AI ​​host didn’t seem bothered but expressed surprise, exclaiming “Woah!” Before politely asking the human to chime in.