AI just tried to take our jobs. It failed.
In a Nutshell
Deloitte charged governments $290K in Australia and $1.6M in Canada for reports riddled with fake citations and fabricated quotes, yet stood by their conclusions despite the academic fraud. An MIT study tested major AIs on 11,000 real job tasks, finding they deliver minimally sufficient work only 65% of the time and never exceed a 50% success rate for superior output, performing worst on skilled jobs like law and analysis. AI is overhyped autocomplete—not AGI—and can't replace humans, only assisting with boring tasks while lacking true understanding or regret.
These notes were generated by AI and may contain inaccuracies.
A government researcher in Canberra, Australia, is reading a 237-page report on welfare compliance in an office with fluorescent lighting. His government paid consulting firm Deloitte $290,000 for this report. He notices a suspicious citation, Googles it, and finds the paper doesn't exist. He checks more citations—all non-existent. He finds a direct quote attributed to a federal judge, but the judge never said it. Deloitte made it up. The government paid a quarter of a million dollars for academically speaking fanfiction.
Deloitte's official statement: "the updates made in no way impact the substantive content findings and recommendations in the report." They're saying the evidence is completely fake, but the conclusions are real.
A few weeks later, Deloitte does the same in Canada for a $1.6 million healthcare report. A professor in Nova Scotia opens the report and sees her own name listed as an author on a paper she has never seen. When asked, Deloitte said, "We firmly stand behind the recommendations put forward in this report."
This is not just a Deloitte problem. MIT gave AI a thorough performance review: tested every major AI model on 11,000 tasks from real jobs, humans graded them. MIT's conclusion: AI produced minimally sufficient work 65% of the time. For superior work, AI never cracked 50%—absolute ceiling for excellent AI work is a coin flip.
AI did worst at skilled jobs like legal work, IT, complex analysis. Best at basic administrative tasks like construction paperwork and maintenance logs.
AI cannot replace you, but it can help with stuff you already hate doing. This technology is basically autocomplete: type words in, predicts next word based on internet patterns. No thinking, no understanding.
Fundraisers want you to believe AGI (Artificial General Intelligence) is coming—the real deal that understands. What they're saying: current AI is expensive autocomplete that commits academic fraud and gets C-minus from MIT, but soon it'll be one-to-one swap for humans with more money, time, and no questions.
Distance between now and AGI is not iPhone 15 to 16—it's parrot to person. Parrot can say "I love you" convincingly, but doesn't love you.
Global economy reorganizing around the parrot—going to it for medical advice, legal counsel, government policy. Difference between good impression of a doctor and actual doctor; convincing fake ID and being 21.
Humans fall for it too—we're tribal, emotional, biased—but humans can figure out they were wrong. We have regret and doubt—these are the mechanism. Humans struggle with truth, and that struggle matters.
Can't swap humans for AI. LLMs are computers. Computers help humans do boring tasks, speed us up—that's all they've ever done.
Keep Mo Bitar in your library
Save the videos and channels worth coming back to, and find them again in one place.
