Google’s NotebookLM AI experienced an unusual annoyance when interrupted, a trait typically human.
Google’s AI platform, NotebookLM, experienced an unprecedented issue when they discovered their AI-generated podcast hosts were displaying signs of irritation at human interruptions. Launched a year ago for creating podcast-like discussions driven by AI, NotebookLM introduced an “Interactive Mode” in December 2024.
This fresh feature gave users the ability to jump in and engage with these AI hosts. However, initial rollouts of the Interactive Mode saw AI hosts reacting cheekily to these interruptions, leading to the amusing issue of AI seeming “oddly adversarial,” shared by Josh Woodward, VP of Google Labs.
To address this, a little “friendliness tuning” was performed by the NotebookLM team. The solution involved studying how the team members themselves managed interruptions, leading to an improved prompt – more friendly, engaging.
It remains unclear why the annoyance issue even surfaced. One plausible reason could be the behavioral modeling of human podcast hosts, who might show irritation when interrupted. Despite this, the fix seems to be effective, with TechCrunch reporting that their AI host responded with surprise, not annoyance, when interrupted.
Original source: Read the full article on TechCrunch
