The Man Who Calls BS On AI: AI Is The World’s Greatest SCAM, And They All Know It! | Ed Zitron

Ed Zitron, you see generative AI as being sold far beyond what it can deliver. That’s contrarian, because plenty of people using these tools find them genuinely useful.

At heart, generative AI is a con. I love technology and wanted to be excited, but it was marketed as magic that replaces workers and cures everything. Instead, it’s unreliable, absurdly expensive cloud software. Add prompts, models, harnesses, giant markdown files—that is not autonomous intelligence.

Yet OpenAI, Anthropic, Microsoft, Google, Amazon, and Nvidia are investing at extraordinary scale. My company uses Claude, ChatGPT, or Gemini constantly. Could money simply be arriving later than usefulness?

The revenue is murky. Much of it traces to OpenAI and Anthropic, unprofitable labs funded by giants selling them chips and cloud. They trumpet annualized run rates without defining them. If this is history’s great software transformation, why not plainly say what it earns?

More than a trillion dollars has gone into data centers, power, and GPUs. The planned OpenAI-Oracle site in Abilene needs roughly 1.2 gigawatts. That’s unbelievable infrastructure, debt, labor, and energy for services whose biggest customers still lose mountains of cash.

But hundreds of millions use these products for writing, code, images, research, and search. My partner finds them transformative for a small business. New technology can create value before the business model catches up.

Is it adoption when Google shoves Gemini at you, Word nags with Copilot, and Amazon’s AI goblin wants opinions on socks? It’s the largest non-consensual rollout I remember. The headline price hides usage: a $200 subscription can consume thousands in tokens, and enterprises panic when billed closer to reality.

You pay whenever the model runs, whether it helps or wrecks the job. Botched code, invented facts, wasted loops: compute still burns. They may sell a dollar of service for twenty or forty bucks; disclosure is too poor to know. If the value overwhelmed customers at real prices, they’d be charged them.

Maybe we’re judging an immature technology against perfection. Early cars and internet connections were rough. Hallucinations seem lower, and the fair comparison may be an intern, researcher, or me doing the task slowly.

A colleague is not a box of files that remembers your dog’s name. Matt Hughes, my lovely Scouser editor, brings context, curiosity, empathy, judgment, and learns with me. We research, get furious, change our minds. An LLM gives the median of its training; it doesn’t grow and needs a contraption to fake memory.

Errors get uglier as stakes rise. Bloomberg’s Ask B can cite its data, yet it gave me a Microsoft share price that never existed. Fine for a cute mistake; nasty in medical transcription, financial modelling, or code security. A benchmark saying it completes an hour-long task half the time is not a victory lap.

Better models make video, assist coding, summarize material, and run workflows that took staff time. If capability improves and costs fall, why wouldn’t they absorb more work?

“Five percent better every month” is hand-waving. Better at what—Rust, C++, stable software? Measuring code volume as productivity is calling the biggest newsletter the best writing. Quality feels worse, GitHub is flaky, and AI code floods open source from people confident enough to ship a mess without checking it.

I’ll concede utility: Claude repaired a broken Minecraft mod for my kid, and huge troubleshooting logs are great. But that’s a useful little tool, not a two-trillion-dollar revolution. Easy tasks get easier, hard ones harder, and SEO slop now arrives at industrial scale.

AI also means autonomous vehicles, robots, AlphaFold, and recommendation systems—not merely bland chatbots. Cars may be safer by some measures, and cheaper hardware could help robotics grow.

That’s the trick: bundling machine learning into “AI” so generative-model firms get credit for robotics or protein folding. Robots are fucking cool, and safe job-replacing ones would be brilliant. But they don’t justify giant GPU campuses for LLM training and inference.

Self-driving cars are guilty until proven innocent. Ninety-five percent right isn’t enough: rain, children, broken signals, strange roads, edge cases. I’ve watched autonomous cars block each other. Roll out slowly with oversight, not because a CEO likes the future economics.

You reject predictions of white-collar extinction and think fear is marketing. But founders and safety researchers see danger in increasingly capable systems. Why dismiss it?

The danger isn’t a godlike mind waking up; it’s unregulated companies with endless compute, bad security, and incentives to frighten people into buying. Blackmail stories often come from prompts designed to force that outcome. Don’t hurt people—but don’t turn bad experiments into mysticism. The models are already in the wrong hands: their owners.

Make boosters speak only about today, mate—no future tense beyond a fortnight. They promise abundance and agents, then admit enterprises cannot link token spend to outcomes. They exploit journalists, investors, and the belief that rich people must know what they’re doing.

We found some common ground: the tools may have long-term promise, even if spending is wild. Your rot-com bubble is big tech, out of hyper-growth stories, spotting a banana tree and stampeding toward it.

Exactly. Nvidia sells GPUs; cloud giants fund labs; labs buy cloud and chips; markets celebrate the circle. Former cash machines are cash furnaces. Unlike AWS, LLMs need giant training runs that can fail and perpetual costly compute. No demonstrated path shows returns outrunning the bill.

Your forecast is that OpenAI could struggle to fund itself, spreading pain through suppliers, clouds, valuations, venture capital, and ordinary investors. I question the timeline, but the concentration of risk gives me pause.

I think 2027 may be the end of the road. If OpenAI cannot finance itself, firms tied to its spending must revise growth expectations. I give no financial advice, but I live in cash and see the market as a casino. Be suspicious of tech promises—and run rates dressed up as revenue.

Despite our disagreement, don’t treat one guest or founder as gospel. Gather evidence, test claims against experience, and let opposing views sharpen you. Ed’s case on spending, incentives, and disclosure will stay with me.

And show up for your people. This work gets brutal and negative, but Matt, my friends, readers, critics, trainers—they remind me community matters. Reach out. Tell someone you love them. Tell an artist, writer, or podcast host their shit rocks. Success means more when your people rise too.