Raf Michalowski's team built an AI media machine producing:
→ 60,000 videos/day
→ videos costing less than 1 cent each
→ millions of uploads/month
And YouTube KEPT approving the channels.
The craziest part? It actually worked.
AI slop isn’t coming. It’s already here.
It always starts small. A simple AI experiment quietly evolves into custom models, massive infrastructure, millions of uploads, and eventually a $50k/month engineering operation.
That’s how AI gold rushes begin.
After 60-plus episodes and six years, I've sharpened my podcast around a single question.
The show used to be a long-form conversation with anyone I respected — Nate Silver, David Epstein, Edward Chancellor, Brad DeLong, Jim Grant, Ben Hunt, Alex Hutchinson, Daniel Negreanu, Ben Sulsky, Haralabos Voulgaris, James Blake. Authors, traders, poker pros, athletes. The variety was the point. There are few greater pleasures than reading a great book and then getting to talk with the author, unrushed, for 1-2 hours.
Now I'm narrowing the lens. It's called The AI Metagame, and it's not about tools or trends. It's about how the smartest operators are playing when the rules keep changing — the patterns, strategies, and second-order effects most people aren't yet seeing.
The latest episode is the clearest example so far. I sat down with Nate Silver and Raf Michalowski on AI-generated content at scale: how one team uploaded 60,000 AI videos in a day, why "AI slop" is becoming a real economic factor, and how LLMs are shifting the bottleneck from execution to ideas.
Watch the full conversation: https://t.co/2us9DohMIC
Explore the series: https://t.co/K9cEuy1LEV
Eventually, most content online will be AI-generated.
This prediction sounds crazy until you hear the economics:
Videos cost less than 1 cent to make, Millions of uploads per month, Fully automated pipelines, Real revenue at scale.
The internet is about to change faster than most people realize.
"LLMs have made products easier to build, but harder to find good ideas."
One of the most important points from episode 5 with @NateSilver538 and Raf Michalowski
Execution used to be the moat, now everyone can build which means you need an edge as the bottleneck shifts.
For now, you still have to babysit AI.
Three places humans remain essential: ideas, taste, and attention to detail.
AI compresses information efficiently but loses acuity in the process. In 10,000+ hours building sports models with Claude, @NateSilver538 caught multiple major bugs.
Expert judgment isn't optional.
Ep 5 of The AI Metagame is live!
I chat with @NateSilver538 and Raf Michalowski about AI slop and the future of the media landscape.
At one point, Raf’s system was uploading 60,000 AI-generated videos a day to YouTube.
Nate works the slice of media where calibrated forecasting and accountable predictions still pay—and AI has barely cracked it.
We talked through both sides — what scales, what survives, what's left for humans when content becomes a pure scale game.
450+ tasks. Constantly reprioritized. Deciding what matters specifically to you.
Most people are overwhelmed by their to-do list.
He outsourced his.
This is the level of leverage LLMs bring.