Monday, December 23, 2024

Google’s AI Turns the Words “Fart” and “Poop” Written 1,000 Times Into an Entire Podcast

Must read

This is hysterical.

Garbage In, Garbage Out

Podcasts are easy targets for parody. Sometimes, they really are just a couple dudes talking into badly EQ’d mics, who may or may not have come prepared to actually discuss their subject at hand with any real depth.

But when it comes to ranting about anything on the spot — and when accuracy is no object — it’s hard to beat AI. A recent example of this comes from a Reddit user, who, using Google’s NotebookLM AI model, decided to find out just how far you could push its “Audio Overview” tool.

The tool creates a podcast script by summarizing your notes. In an experiment spotted by Ars Technica, the Reddit user fed the tool an entire document that is just the words “poop” and “fart” repeated one thousand times.

And lo and behold, the pair of AI hosts end up doing an admirable job of riffing on the crappy document, down to imitating the natural, “um”-filled cadence of podcaster speech. There’s even plenty of faux-profundity to go around — a hilarious display of the ability of large language models to literally bullshit their way through anything.

“One of the first questions we have to ask ourselves is, can we find meaning in the seemingly meaningless? Because that’s what we’re faced with here,” one of the AI hosts asks.

Budding Fartists

Well, one thing’s for sure: the AI is very good at talking out of its ass.

Right off the bat, one host compares the poop-fart document’s heavy use of repetition to Andy Warhol’s painting of soup cans, as an example of how our minds like to seek out patterns.

“And those things can be surprisingly powerful, can’t they?” the AI asks. “It’s like our brains refuse to believe it means nothing.”

The AI even brings up the absurdism of the Dada art movement, invoking Marcel Duchamp’s readymade sculpture of an upside-down urinal.

If you listen to the whole, nearly ten-minute clip, you’ll eventually be treated to sagacious aphorisms like, “Sometimes a poop is just a poop and a fart is just a fart.”

Bridge Too Far

Of course, bullshitting has its limits. No, there isn’t a “really valuable lesson” here about how “we approach the world around us.” Didn’t you guys say that a poop is just a poop?

There’s no point in trying to find actual nuggets of wisdom. But on the surface level, it’s very convincing stuff, with the right amount of casual humor you’d expect on a podcast, that, at times, verges on being unnervingly human.

“We’re meaning-making machines, aren’t we?” one of the hosts says at one point.

Let’s hope those weren’t the words of some buried, nascent AI consciousness breaking through.

More on AI: There’s a Problem WIth AI Programming Assistants: They’re Inserting Far More Errors Into Code

Latest article