when will they finally see?
march 16, 2025
it was the autumn/summer of 2022. an 11-year-old kid was bored, looking for an ai that could tell stories. i stumbled onto openai's playground and saw a selector bar on something named text-davinci-002. i had no idea what that meant, but i started typing.
one of my first prompts was about "a boy and his robot friend having bubble tea, and the robot reacting to it." gpt-3 generated a story where the robot started to understand why humans liked bubble tea—it "grew on him." (word by word from gpt-3, stil remember it) as a kid, that felt like magic.
the computer painted.
around the same time, i started using dalle-2. it couldn't make any realistic images back then, but it made really good & tasteful oil paintings and abstract art. again, the kid took it to my parents, friends, my whole family. their reaction?
"huh, so ai can draw now. interesting."
the best birthday gift
fast forward to late november 2022, right before my birthday. my dad, knowing i liked tech, told me about something new called chatgpt that was all over the news. i tried it out, and that was the best birthday present i ever got.
i don't remember my first prompt, but it was probably something like "who are you?" chatgpt responded with a paragraph about the nature of human existence. i showed my parents, my friends, my whole family. they just said, "oh, cool, a chatbot that spits out text."
but for me? it was the moment i started learning to code again (after learning a bit of python and stopping at 10).
the gpt-4 moment
march 14, 2023. the gpt-4 announcement stream.
i watched as greg brockman sketched a website on a napkin, and gpt-4 turned it into a working website. that was the moment my perspective on ai shifted. before, i thought of it as a really powerful tool. but after that? i genuinely believed agi was coming soon.
i subscribed to chatgpt plus, started coding with gpt-4, and it felt unreal. sometimes, i'd copy text from documentation, paste it in, and gpt-4 would give me a working program. it felt like magic.
again, i showed my friends and family. again, they shrugged: "it's just a program that makes text. and now it can look at images. cool, i guess."
they finally noticed
it wasn't until gpt-4o that people around me really started using llms. my mom started asking it about her daily tasks, like taking a picture of food and checking how healthy it was. she loved the internet search feature, often using it to look things up.
finally, people were using it.
a few months ago, i showed my aunts gpt-4o. their first request? "make it write a dramatic soap opera script" i was speechless. i showed them a literal superintelligence, and they wanted soap operas.
but since gpt-4o was kinda "sloptimized" back then, it couldn't generate good dramatic scripts. after a while, they just shrugged:
"eh, i don't see why this ai thing is that useful after all."
o1 and onwards
right now, when i code, o3-mini-high and claude feel like superpowers. projects that once seemed like someday-maybe ideas have turned into afternoon builds. but when i showed my family gpt-4o, o3-mini-high, and claude-3.7-sonnet side by side, my mom just said:
"what the diffrence?"
that's when it hit me—the problem isn't that ai isn't smart enough. it's that most users aren't technical enough to see the difference. it's easy to tell a kindergartener from a high schooler, but most people can't distinguish between a high schooler and a PhD.
we're heading toward a future where maybe 0.1% of people will be able to tell the difference between a cutting-edge model and an older one. i saw it when lmsys started struggling with model evals—we just aren't smart enough anymore.
honestly, openai could swap gpt-4o for an 8b model tomorrow, and outside of techical fields / circles, no one would notice with their "chatgpt rewrite email make look better and more formal" requests.
i saw the ripples before the wave. i screamed and screamed for people to get their boards. most of them just stood there laughing.
when it did hit?
a couple people surfed, most just got drenched
