Google’s NotebookLM had to teach AI-powered podcast hosts not to be annoying to people

Prevention is annoying. Apparently even AI-generated podcast hosts agree.

Or so Google NotebookLM users discovered. NotebookLM started last year and it went viral thanks to its podcast-like discussion-generating feature created entirely by AI from user-uploaded content, the conversation is moderated by AI bots acting as podcast hosts. In December 2024, NotebookLM launched a new feature called “Interactive Mode”. allows the user to “call in” the podcast and ask questionsartificial intelligence interrupts their owners while talking.

When the feature was first introduced, AI owners seemed annoyed by such interruptions. Josh Woodward, vice president of Google Labs, explained to TechCrunch that they occasionally gave short comments to human callers like “I made it” or “just like that,” which was “weirdly adversarial.”

So the NotebookLM team decided some “friendly tweaking” 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 learning how to have its members respond more politely to the interruption.

“We tested different proposals, often learning how people on the team would respond to outages, and landed on a new direction that we thought felt friendlier and more engaging,” he said.

It’s not entirely clear why the problem occurred in the first place. Human podcasters sometimes show frustration when interrupted, which can result in the system’s training data. A source familiar with the matter said the case likely stemmed from the incentive design of the system rather than the training data.

Regardless, the fix seems to be working. When TechCrunch tested the Interactive mode, the AI ​​owner wasn’t annoyed, but surprised and said “Wow!” before asking the person to sound polite.