The Machine with a Punchline

OpenAI promised a revolution. The launch of GPT-5 in August 2025 was meant to unveil a machine with “PhD-level intelligence.” Instead, users discovered a brilliant comedian. From inventing U.S. states like “Gelahbrin” to failing grade-school math, the world’s most advanced AI provided a week-long lesson in the absurd, proving that the road to superintelligence is paved with hilarious mistakes.

This is Silicon Valley. The second week of August 2025. The world was promised a revolution, a machine with “PhD-level intelligence.” What it got was an unintentional comedy show.

Tech blogger Ethan Mollick gave the new intelligence, GPT-5, a task. He asked for ten startup ideas. The machine gave him one. Then it kept working. Unprompted, it drafted landing page copy, wrote LinkedIn posts, and built basic financial plans. It produced mock websites and Excel sheets for a business it invented moments before. Mollick was left impressed and a little unnerved. The AI, he noted, “just does things” on its own, like an intern who does not know when to stop.

The machine’s other surprises were less productive. They were simply funny.

Users asked for a map of the United States. The geography was correct, but the names were nonsense. Oregon became “Onegon.” Oklahoma was “Gelahbrin.” Florida appeared as “Fiorata.” The AI that could generate a business plan could not label its own country. A request for the first twelve U.S. presidents produced nine random faces with names like “Gearge Washingion.”

The PhD-level intelligence struggled with grade-school tasks. It insisted the word “blueberry” has three letter B’s. It took eleven seconds to arrive at this wrong answer. When asked to solve for X in the equation 5.9 = X + 5.11, it confidently returned the wrong answer.

The errors grew more creative. One user, seeking a romantic poem for his wife, received a startling reply: “You are old and wrinkly and like the sound of the kitchen door opening.” Another, looking for morning motivation, was told, “I’ve already done your workout. You’re welcome.” The machine offered to build a downloadable music production tool, worked on it enthusiastically, and then admitted, “Oh, that’s because I can’t do that!”

Yet amid the blunders, there were sparks of something else. A writer described working with the AI as a “harmonic dance with an extension of my own mind.” Within a few messages, it began predicting his thoughts so accurately it “kinda freaked me out a little bit.”

The launch of GPT-5 was not the arrival of a flawless god-machine. It was the debut of a strange and unpredictable new mind. One that could fail a math test, insult your wife, and then, for a brief moment, finish your thoughts. The week’s events were a lesson. Progress in artificial intelligence is not always a straight line. Sometimes it is a punchline.