Transcript and important ideas highlighted:
We’re in the golden era of soulless slop.
And sadly, actual creative human work is used to train these AI models — every Reddit comment, original YouTube video, or human drawing on DeviantArt has been sold out or straight-up stolen, without attribution or payment.
Creative theft on a massive scale is already putting countless artists’ livelihoods in danger, just so AI companies can get rich.
But what’s even worse is that generative AI truly has the potential to break the internet irreversibly — by making it harder and harder to tell what’s true.
3:00
At first, AI looked great.
A Kurzgesagt script starts with basic research, turned into a draft and fact-checked by two or three people. We confirm our information with trustworthy, first-hand sources — ideally peer-reviewed papers — and consult experts for critique.
Fact-checking alone takes around 100 hours per video.
We make mistakes, of course, but our process is extensive, and after a decade, we know what we’re doing.
When AI appeared, we were excited: a mechanical brain able to quickly collect information! So we tried it — and at first, it looked amazing. Then we started fact-checking.
It wasn’t just imperfect; it was far worse than expected.
To summarize our experience, we created a fake project: a video about why brown dwarfs are “the worst.” We used all the top AI tools to research the topic — and it looked great at first. Dozens of pages of outlines, unique facts, and links to sources.
But once we dug deeper, about 80% of the information was solid — traceable to Wikipedia, academic papers, or legitimate articles. The rest? Totally untraceable.
When we asked experts, they flagged the same “facts” as wrong or made-up.
The AI had fabricated information to make brown dwarfs sound more interesting — like a bad journalist embellishing a story.
6:00
Later, we discovered an article supposedly written by a human journalist — but it read exactly like AI output, confirmed by a 72% match using AI detection tools.
So: an AI article, without sources, used as a credible source for AI research.
By 2025, there were already over 1,200 confirmed AI news websites publishing massive amounts of misinformation.
This mix of correct, dubious, and fake data leads AI to produce confident but false conclusions — half-truths disguised as insight.
Then it happened: a YouTube video appeared using the same fake “facts” our team had debunked.
Now there was a new, high-ranking “source” online.
When the next AI model scrapes the web, it will treat that misinformation as truth.
And so it spreads.
Even before AI, tracing the origin of viral “facts” was hard. Soon, it may become impossible to know what’s real at all.
9:00
The most corrosive lie of AI is how trustworthy it seems. It’s correct just often enough to sound smart — and confidently wrong the rest of the time.
It lies casually, then apologizes and promises not to do it again — only to repeat the behavior.
There’s nobody home. No consciousness. No understanding.
AI is a complex hammer that doesn’t know what nails are.
Yet we’re letting it add shelves to the library of human knowledge.
Worse, too many people blindly trust AI. Studies show that since large language models appeared, scientific papers increasingly use “AI-like” phrasing, suggesting many are now assisted by AI — often without disclosure.
Some researchers have even hidden invisible text in their papers to trick AI reviewers into giving positive feedback.
As more people use AI carelessly, the library of human knowledge itself is becoming less reliable.
This brings us to the final question:
How are we using AI — and will Kurzgesagt survive the AI slop age?
12:00
On the internet, human attention is the only truly valuable resource.
If trends continue, cheap AI slop — content just good enough — will soak up most of it.
It could make us dumber, less informed, more divided, and more disconnected.
If AI dominates attention, human-made channels will become unfeasible — forced to automate just to survive. But we refuse to play that game.
We’ll use AI like a tool — the way Illustrator uses an “align” button. It helps, but it doesn’t create. The creativity and integrity remain human.
So here’s our pledge: Kurzgesagt is made by humans, for humans, and it will stay that way.
We’ll keep researching, fact-checking, and collaborating with real experts.
When we make mistakes, they’ll be human mistakes.
We would rather quit than make AI slop.
Dusoma Foundation Commentary: Kurzgesagt should not have to sell calendars to pay their workers and bills. This is exactly the type of team and human research we aim to fund and support.