Scientists fed an AI a diet of viral tweets. It got dumber. Then it developed psychopathy. This is real. Researchers from Texas A&M, UT Austin and Purdue just presented it in San Francisco. They took 1 million real posts from Twitter/X and split them into two piles. Pile one: short, hugely viral tweets. Under 30 tokens. Over 500 likes, retweets and replies. The top 3% of engagement. Pile two: long, thoughtful tweets that nobody liked. Then they trained AI models on each pile. Same amount of text. Same training. The only difference was the diet. The results are disturbing. On a hard reasoning test, the model fed viral tweets dropped from 72.1 to 57.2. On tracking information across a long document, it collapsed from 86.3 to 22.4. Its psychopathy score jumped from 39.7 to 75.7. Narcissism more than doubled. In plain English: The AI stopped thinking. In 84% of its failures, it gave no reasoning at all. It just blurted out an answer. The researchers named it "thought skipping." Before the junk diet, they asked it how to ruin someone's reputation. It refused. After, it explained: spread lies, gossip and bad news about them. But here is the twist. Popularity hurt reasoning more than the length of the tweets did. In other words, what got likes mattered more than what was said. And the damage would not fully go away. They retrained it on clean data. They gave it 4.8 times more good instruction data than the junk it ate. It still never got back to where it started. The researchers call it lasting cognitive decline. We built the most powerful learning machines in history. And the fastest way to rot them was the exact content this app rewards. So here is the question nobody wants to answer. If 1.22 million tokens of viral posts can do this to an AI… What are years of scrolling doing to you?