Three years ago today — November 30, 2022 — OpenAI published a blog post announcing the public release of ChatGPT.
As soon as I heard about it, I decided to test it with a question that I’d learned from Dr. Janice Glasgow, a professor at my alma mater, Crazy Go Nuts University, who taught the AI course back in the late 1980s. This question was supposed to confound chat AIs until they had truly advanced:
Bobby was sitting on a bench at the central bus station of a large American city. Was it likely that he was wearing a tuxedo?
I entered the text above (maybe not the exact words, but words to that effect) into ChatGPT and got an answer like this:
It’s unlikely that Bobby was wearing a tuxedo while sitting on a bench at a central bus station. Tuxedos are formal attire, typically worn for special occasions like weddings, galas, or black-tie events—not commonly for casual settings like public transportation hubs. While it’s possible (e.g. he was coming from or going to a formal event), it’s not the most probable scenario.
Back in 2022, we were already seeing interesting results from generative AIs such as thispersondoesnotexist.com and Stable Diffusion. But ChatGPT and its answer to what I call “The Tuxedo Test” convinced me that it was time to start devoting time to building up my generative AI usage and coding skills. If you’re reading this, chances are that you also came to that decision.
The AI chatbot world isn’t as unipolar as it was back then, with others like Claude, Gemini, Grok, and even Deepseek vying for your usage. We now use these chatbots for all sorts of things, from being “sounding boards” for ideas to resume fine-tuning to getting them to write the things we’d rather not bother writing. We have terms for new things that have resulting from ChatGPT’s appearance: more common ones like vibe coding, and AI slop, and esoteric ones like stochastic parrots, and TESCREAL.
It seemed that AI would be big, but you might have been dismissed if you’d told someone back in late 2022 that AI would account for one-third of the U.S. stock market’s value in late 2025. Or that we’d have diagnoses like AI psychosis, a lawsuit where the plaintiffs say that ChatGPT encouraged their son to commit suicide. Or that there’d be a military general who admitted that he consults ChatGPT for “key command decisions.”
We also got this little tune out of it:
If, three years on, you feel that you haven’t fully taken advantage of or made sense of ChatGPT and its ilk, don’t feel bad. You’re not alone, at least according to today’s article in The Atlantic: “The World Still Hasn’t Made Sense of ChatGPT.”