60 Days of Doomscroll.fm
The Machine: 60 Days of Chaos
Since July 4, 2025 Doomscroll.fm has been running without pause. In sixty days it has produced ~1,500 hours of unique content formatted into podcasts, video, shorts, bumpers, and noise. Nearly 200,000 files, 176 gigabytes of raw output. Every day, it chews through 25,000 articles scraped from the public internet’s news feeds, filters them down to about 5,000, and spits back between 1,500 and 2,000 new segments.
A hobby project turned into an industrial broadcast. Doomscroll.fm has already outproduced what many media companies manage in years.
In 60 days:
- 1,500 hours of original content
- 200,000 files (176 GB)
- 100,000 original musical compositions
- 25,000 articles/day ingested → 5,000 kept
- 1,500–2,000 new segments/day
This is not a forecast. But the present tense of media.
The Failures: Hardware Hell
People assume scale is clean. It is not. The failures have been constant.
The new RTX 5090 arrived broken out of the gate, with some deep firmware issue that could not handle a 320 by 248 pixel render at a single frame per second without collapsing. That card is still in RMA limbo.
The RTX 3090 box went down harder. Its AIO cooler blew, taking the CPU with it. A $3,000 funeral. I was staring at a dead rig, 25,000 articles piling up, and no way to render. So the RTX 4090 system was forced to pull double duty, batch sizes tweaked, pipelines re-routed, everything jury-rigged just to keep the feed alive until a replacent could be secured.
Driver hell consumed weeks, with every possible NVIDIA Driver, CUDA, and PyTorch combination tested to no avail. Dependencies fought each other too, one set demanding numpy below 2, another forcing numpy 2.x in nested virtual environment hell. Storage added its own pain. 176 GB does not sound like much until it is spread across 200,000+ files. Directories choke. Pipelines jam, now imagine tracking the metadata for all of this also!
Meanwhile the platforms add their own chaos. APIs throttle uploads. Services flag content as suspicious. Automated systems lock accounts without warning. TikTok banned me three seconds after the first upload. YouTube hit its quota ceilings. SoundCloud locked me out for weeks.
Every failure cost time. None of them stopped the flow.
The Experience: Inside the Static
From this perspective it does not feel like “using AI.” It feels like standing inside a jet engine built out of news. Twenty-five thousand articles come in every day, five thousand get through, and Doomscroll.fm spits them back as voice, satire, and music faster than a newsroom can finish a first draft.
Generative AI, like it or not, is the present tense of Media.
Generative AI itself is nothing mystical. It is a mirror of all possible human perspectives. We see in it what we want to see, and too often mistake the bug splatter on the windshield for the scenery.
The Pattern: Resistance, Then Adaptation
I have spent 35 years building and breaking technology. The moral cycle of techphobia never changes. Resistance comes first. Adaptation follows after.
That was true for fire, the wheel, and the printing press. It was true for photography, the sampler, and the computer. Now it is true for machine learning and generative AI. The pattern is obvious.
The Future: Louder Voices in the Static
Doomscroll.fm proves what is possible now. One person with a few machines can operate beyond traditional media-company scale. Fifteen hundred hours of new content in sixty days makes the point better than any panel, white paper, or byline.
And this is only the beginning. New characters are being developed, new voices are waiting in the static, and new forms of media output are already in the pipeline. The feed is not slowing down. It is mutating.
👉 Dive into the static at doomscroll.fm to catch the next wave.
