Why AI Video Feeds Are Full of Destruction: A Thread on Catastrophe Content

On October 13, AI creator @CurieuxExplorer posted a thread examining a phenomenon: AI video feeds are saturated with catastrophic destruction imagery—earthquakes, tsunamis, missile strikes on cities—that routinely racks up millions of views. While he creates with AI himself and marvels at its capabilities, he questions why creators flock to the same spectacle of destruction, rather than blaming the technology or assigning responsibility.

Confirmed

• The author cites a 2024 Nature study showing that repeatedly training models on AI-generated output without adequately preserving real data leads to "model collapse"—rare patterns vanish, errors accumulate, and diversity degrades.
• One study analyzing 95,282 news articles and 579 million+ social posts found that news links with negative content were 1.91x more likely to be shared.
• Another study in Nature Human Behaviour, analyzing news headlines with 370 million+ impressions, found that each additional negative word in a headline boosts click-through rate by roughly 2.3%.

Unconfirmed

• On whether "mass-producing disaster videos amounts to teaching AI destruction," the author is cautious: generated videos do not necessarily enter training data and won't automatically have an impact; but if synthetic content does enter future datasets, it will shape what future models learn.

Why it matters

• The author traces an attention feedback loop: creators chase attention, audiences reward content that shocks them, platforms learn what best sustains engagement, and creators then replicate what works. Citing research that positive feedback on social media reinforces future expressions of outrage, he suggests the spectacle of "visual destruction" in AI video may form a similar creative loop.
• On responsibility, he argues the blame for the explosions and destruction filling AI video lies not with the tool but with everyone in the chain—creators, viewers, platforms, developers training and deploying the models, and institutions deciding their uses—and extends the discussion to autonomous weapons that can act without human intervention once activated.

2026-10-13 ~ 2026-10-13 · 8 related posts

• [source] AI Can Imagine Anything, So Why Do Its Videos Keep Destroying Cities? — CurieuxExplorer · 2026-10-13
• The Creator's Dilemma: AI Video Feeds Are Stuck on Spectacular Destruction — CurieuxExplorer · 2026-10-13
• Each negative word in a headline boosts CTR by ~2.3%, Nature study finds — CurieuxExplorer · 2026-10-13
• [source] Study: Negative News 1.91x More Shareable, AI Video Tools May Amplify Harm — CurieuxExplorer · 2026-10-13
• The attention feedback loop: how creators, audiences and platforms reinforce spectacle — CurieuxExplorer · 2026-10-13
• Are We Training AI on Destruction by Mass-Generating Disaster Videos? — CurieuxExplorer · 2026-10-13
• [source] Nature 2024: Repeated Training on AI Output Causes 'Model Collapse' — CurieuxExplorer · 2026-10-13
• Who's Responsible for AI's Destruction Content? The Answer Isn't the Machine — CurieuxExplorer · 2026-10-13