Have you ever been interrupted mid-sentence? Of course you have — and it's annoying. But thanks to Google, at least AI will start to share our pain.
Last year, Google launched NotebookLM, a service that allows people to create AI-generated podcast conversations. In December, NotebookLM added their latest feature: Interactive Mode! This allowed users to "call in" to the podcast … but the AI hosts reacted about as well as a 300-pound man with purple hair being misgendered in a Portland Starbucks.
AI hosts were as annoyed as we are when we're interrupted, with responses like "I was getting to that," or "As I was about to say." It's like Karine Jean-Pierre on steroids!
Josh Woodward, VP of Google Labs, told TechCrunch this was "oddly adversarial," with NotebookLM announcing they're working on some "friendliness tuning."
If only the Terminator had some friendliness tuning …
Woodward explained the fix involved testing prompts — basically feeding the AI scripts on how to handle interruptions without snapping.
"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," he said.
Nothing like forcing engineers to code politeness into the AI that's probably going to replace them in a few years! "Oh, sorry I took your job!"
Funnily enough, no-one really knows why the AI went full Karen to begin with. Was it bad training data, like audio of human podcasters losing their cool? Or was it just the way the system was designed?
But in the end, does it even matter? Honestly, if we're going to be murdered en-masse by robots, I don't want them apologizing for it.
Follow Ian on Substack or X (@ighaworth).
P.S. Now check out our latest video 👇
