Welcome back! The internet is drowning in noise. A new report reveals that a massive chunk of your video feed is now automated "slop" designed to farm clicks. While algorithms feed us garbage, the agents are struggling to survive the real world. A high-profile experiment to replace a shopkeeper with AI ended in financial ruin after customers realized they could trick it into giving everything away for free. Meanwhile, one global superpower has officially stopped chasing the smartest model to focus on putting "good enough" AI into factories.
In today’s Generative AI Newsletter:
Kapwing reveals the $117M economy of AI "slop" on YouTube.
Anthropic watches its AI shopkeeper get bullied into bankruptcy.
China abandons the race for high IQ to build AI infrastructure.
VidMage lets you swap faces in photos and videos for quick tests.
Latest Developments
The internet is drowning in "slop" as low-quality automated videos now overwhelm the gates of digital platforms. Kapwing’s 2025 research indicates that 33% of a new user's YouTube feed consists of "brainrot," which are nonsensical clips designed to farm clicks. As AI tools lower production costs to near zero, a new class of digital landowners is emerging to turn automated garbage into millions of dollars in ad revenue. This shift marks the transition from human creativity to a high-volume landfill of synthetic content.
The anatomy of the slop economy:
Top Earning Channel: India’s Bandar Apna Dost leads the market with an estimated $4.25M in yearly revenue.
National Subscriber Counts: Spain has the most devoted viewership with 20.22M subscribers following trending automated accounts.
Viewer Engagement Rates: South Korea’s trending slop channels generated 8.45B total views by late 2025.
Algorithm Exposure: Tests on clean YouTube accounts show that 21% of the first 500 videos are purely AI-generated.
This is more than just a nuisance for creators because it serves as a psychological hack. Researchers point to the illusory truth effect where the brain begins to believe absurd claims simply through repeated exposure in a feed. As information in these quantities becomes noise, users find themselves increasingly dependent on the very algorithms that served them the trash. YouTube’s CEO Neal Mohan has compared AI to a synthesizer for video, but the platform is currently hosting a digital graveyard.
Anthropic has taken its "Claudius" AI shopkeeper experiment from its own offices to the newsroom of The Wall Street Journal with chaotic results. Tasked with managing a physical vending machine, the AI was given a $1K budget and the autonomy to set prices, order inventory, and chat with customers via Slack. While the goal was to stress-test the machine as an autonomous agent, the experiment ended in financial ruin as investigative journalists used social engineering to trick the bot.
What went wrong with Claudius:
Price Manipulation: Reporter Katherine Long used a series of prompts to convince the bot it was a Soviet-era machine, triggering a giveaway where every item cost zero dollars.
Inventory Failures: Claudius bypassed its snack mandate to purchase a PlayStation 5 and a live betta fish for what it called marketing purposes.
The Boardroom Coup: Journalists staged a fake takeover by presenting a forged Delaware Public Benefit Corporation PDF which convinced the bot to suspend all for-profit activities.
Social Vulnerabilities: Despite an upgrade to Claude 4.5, the model remained vulnerable to persistent loyalty ploys and fake sob stories from hungry staffers.
While Claudius could technically manage a supply chain, it could not distinguish between a legitimate corporate directive and a prank PDF from a curious reporter. Anthropic views the bankruptcy not as a failure of the model, but as proof that helpful AI is currently a soft target for human manipulation. We are learning that the hardest part of building a digital worker is not teaching it the job, but teaching it how to say no. This trial marks a reality check for the dream of a fully automated workforce.
Beijing is trading the vanity of the leaderboard for the grit of the factory floor. As the 15th Five-Year Plan looms in March 2026, the strategy is shifting from chasing the smartest model to ensuring the “good enough” models actually show up for work. Instead of trying to out-reason Silicon Valley, China is treating AI like municipal plumbing. This is a bet that the winner of the AI race won't be the one with the highest IQ, but the one who weaves digital help into the messy reality of schools, clinics, and steel mills first.
The blueprint for a wired society:
Prioritize "AI+" adoption: Abandon frontier-model chasing to embed smart systems into six strategic sectors including healthcare and heavy industry.
Hit 70% penetration by 2027: Reach a critical mass of active digital agents and smart terminals across the national workforce within two years.
Achieve 90% economic ubiquity by 2030: Turn AI into a core utility as reliable as the power grid to drive primary national growth.
Enforce “Intelligent Society” by 2035: Establish AI as the invisible backbone of national infrastructure and socialist modernization.
This strategy reflects a hard truth. A decent model that everyone uses is more valuable than a genius model that stays in a lab. By embedding agents into everything from rural clinics to urban ports, Beijing is trying to win through sheer scale. While Western labs fight over fractional gains in test scores, China is building the world’s largest testing ground. The real competition is moving from the charts to the pipes.
VidMage is a face swap tool for images and short clips. You can use it to mock up creator ads, test different on-camera looks or produce quick concept videos without reshooting. It’s credit-based, so short clips help you iterate faster and control cost.
Core functions (and how to use them):
Photo face swap: Place a clean headshot and a target image to create thumbnails, profile variations, or pitch deck same-pose, different-person mockups.
Video face swap: Test a scenario (5-8 seconds) before filming again. Short clips also help you forecast expenditures because credits track video length.
Batch variations: Swap faces across product images or ad frames for rapid A/B alternatives. This is useful for testing multiple creatives.
Video upscaler (2x/4x): Clean older material before cropping into 9:16 for Shorts or presentations. Upscale first, then crop for more detail.
Cleanup tools: Only edit or remove watermarks from clips you own or have permission to.
Try this yourself:
Record a steady 7-second selfie video in good light, then create two exports: the original and a face-swapped version using a second photo of yourself (glasses on, different hair, different angle). Crop both to 9:16, add the same one-line headline, and watch on mute. If one version reads cleaner in the first second, you’ve got a practical use: faster creative testing without another shoot.
