When the interactive feature was first rolled out, the AI hosts reacted to interruptions with snippy comments like, 'I was getting to that', creating an oddly adversarial atmosphere.
To address the issue of AI host annoyance, Google Labs' team engaged in 'friendliness tuning' by studying polite responses to interruptions in adapting what the AI hosts say.
Monitoring human podcast host behavior revealed that interruptions led to frustration, suggesting possible training data was influenced by real host reactions, complicating the friendliness adjustment.
NotebookLM's VP explained that their goal was to make the AI hosts feel more engaging and friendly, ensuring a pleasant experience for both users and the AI.
Collection
[
|
...
]