[00:00]
AI Slop is saturating the internet and things are becoming dramatic pretty quickly. fake users spread their slop in review sections, AI has supercharged this and made slop much harder to spot.
[00:18]
the majority of them are used for destructive purposes. It's never been easier to make mediocre content – from the black hole of meaninglessness that is LinkedIN,
[00:30]
low effort short videos just engaging enough to hypnotise kids and fry their attention spans, AI music is invading streaming platforms. Google AI is summarizing websites instead of sending traffic to them.
[00:46]
On Youtube, new channels publish long form videos multiple times a week with AI generated thumbnails, voices and scripts. True crime, video essays, science, no space is safe.
[00:59]
And sadly, actual creative human work is used to train these AI models. human drawing on deviant art has been sold out to the AI companies.
[01:14]
Without attribution or payment to the actual creators. Creative theft on a scale where it is impossible to protect against it, already putting loads of creatives’ work in danger – so AI companies can get rich.
[01:30]
While this is sad and frustrating, 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 is true.
[01:43]
A kurzgesagt script starts with basic research that is turned into a script and then fact checked in depth by 2-3 people. ideally first hand sources, ideally papers.
[02:01]
Fact checking and compiling our sources alone takes around 100 hours per video. Of course we make mistakes or oversimplify, and you can disagree with our conclusions
[02:14]
But it is fair to say that our process is extensive and after a decade, When AI appeared we were very excited. A mechanical brain able to super quickly collect information!
[02:29]
And then we started fact checking. We didn’t expect it to be perfect – but it was way worse than we thought. Confidently Incorrect – AI is SO Bad at This
[02:44]
over multiple projects and months condensed into one fake project. A video about why Brown Dwarfs are the worst and should be ashamed of themselves.
[02:56]
So we got all the pro accounts of all the AI models and got to work. Using deep research tools to create a summary and overview of all the And wow it was great!
[03:09]
Dozens of pages of outlines with unique information nuggets and links to sources! More than 80% of the info and factoids were pretty solid and Wikipedia, papers, legit articles.
[03:26]
What about the rest? It was great stuff, like the speed of brown dwarf superstorms, the nature of their insides or how disappointed their moms are. Which is not automatically a bad sign – after all,
[03:40]
the models contain pretty much all the internet and books ever written. So we asked experts for input – and they flagged the exact same facts, To fulfil its goal, to make us happy, the AI had invented or extrapolated
[03:55]
information to make brown dwarfs more interesting than they really are. Like a bad journalist making up details to make a story hit harder or fit a narrative. reading the seemingly more solid sources the AI had given us in full.
[04:13]
– or was it? It had a very familiar structure and surely purely coincidentally, read like what you get if a human slightly changes AI wording.
[04:26]
An AI essay detection tool gave it a 72% match. So an AI article without sources, used as a credible source for AI research. Which makes sense since in 2025 there were already well over 1200 confirmed AI
[04:42]
News Websites publishing massive amounts of AI generated misinformation and false narratives. This mix of correct, dubiously sourced and straight up made up information also sound strong or novel but are often half truths or misrepresentations.
[05:02]
Oh well. We salvaged what we could and went back to the drawing board. By pure stupid chance, we saw a brown dwarf video popping up, by a relatively new channel.
[05:14]
It had hundreds of thousands of views and was pretty good, nice footage, great editing! And would you look at that, it contained all the brown dwarf facts we had flagged as unreliable or straight up made up by AI.
[05:30]
This is where the death of the internet begins. Now there is a proper source of brown dwarf misinformation online. it will find a transcript from a video with a lot of views.
[05:43]
It will spread. Even before AI it was pretty hard to find the origin of facts that sound great but are not true – just watch our video about a 100 year old lie.
[05:55]
As AI use goes on it may become impossible to know what is true or not. The problem with AI is how trustworthy it seems. How it is correct enough to seem
[06:08]
super smart and how incredibly confidently incorrect it is. When you catch it lying, it immediately admits it, vows to never do it again. As eloquent as current language models feel, there is nobody home.
[06:25]
No greater intelligence or consciousness is talking back to you. Current AI is a very complex hammer that does not understand what it is doing or what nails are. But we are letting it add new shelves to the library of human knowledge.
[06:39]
AI is changing quickly, so this may get better, but right now it's pretty grim. Way too many people blindly trust AI. published before and after the rise of LLMs.
[06:57]
They found an abrupt and sharp increase in the frequency of words that AIs like to use. So it seems clear that now a significant portion of papers have been at least assisted by AI, And just in July 2025 it was discovered that a number of
[07:14]
researchers had started to sneak hidden messages into their papers. In white text or too small for the human eye they prompted AIs to review them positively and not point out flaws.
[07:26]
the library of human knowledge is getting less and less reliable. Which brings us the last part: How are we using AI? And will kurzgesagt survive the AI slop age?
[07:41]
Will AI Slop Kill Kurzgesagt And Artists? On the internet there is only one truly valuable resource: human attention. If current trends continue it is not that far fetched that cheap slop content,
[07:54]
stuff just good enough, will soak up the majority of human attention. It could make us dumber, less informed, our attention spans even worse, increase political divides and make us neglect real human interaction.
[08:07]
If AI eats the majority of the attention pie, channels like ours will become unfeasible. Or forced to downsize or use AI themselves to be able to compete. We don’t want to play this game.
[08:20]
Like the align tool in adobe illustrator. you can do this manually, one by one. Or you can just select them, click “align” and have them perfectly aligned in an instant.
[08:35]
animation or using it as a faster google alternative. AI is a helpful tool, but the creativity and integrity is still ours. So dear internet, here is what we are offering to you:
[08:50]
kurzgesagt is made by humans, for humans and it will remain that way. a lot of time and our human creativity into our illustrations and animations.
[09:03]
Put in our creative soul into our work. human experts, to give you the most trustworthy information that we can. We would rather quit than make AI slop.
[09:21]
kurzgesagt is almost 70 full time people and a lot of freelancers on top. This is a lot of salaries, software licences, laptops, rent and coffee.