When GPT Fails, You Might Be Onto Something
Frustration with AI might be a sign of brilliance. LLMs echo the past—your job is to invent the future.
AIPHILOSOPHYCULTUREDESIGN
Jared Lukes
3/24/20252 min read
Your Favorite Robot Doesn’t Get You—Good
There’s this magic moment that happens when you’re banging your head against GPT and it’s just… not getting it. Feels like failure, right? That creeping doubt that maybe you’re not explaining yourself well enough, or worse, maybe you’re just not making any damn sense.
But hold on. That might actually be the best news you’ve had all day.
Just a Giant Next-Word Predictor
Let’s zoom out for a second. LLMs, these “smart” models we’re all cozying up with, they’re not little people in your computer. They’re statistical beasts. Pattern-matchers. Math-fueled parrots. They’ve digested more human language than any monk in any mountain has ever dreamed of.
And what do they do with all that info? Predict the next word. That’s it. You type “The cat in the…” and they go, “Oh hell yeah, ‘hat’—slam dunk.” Because they’ve seen it a thousand times before. They don’t understand cats or hats. They just know what usually comes next.
The Past, the Past, the Past
These models are experts on what has been said. Not what will be said.
They’re great at answering questions because so many of our questions have already been asked, a million times, by a million mouths. So the LLM doesn’t think. It just finishes your sentence based on what’s statistically likely. It’s just closing your open parentheses with a soft little tap of consensus.
When It Fails, Maybe You’re NOT
So what happens when it doesn’t finish your sentence? When you throw it a question and it fumbles, starts talking nonsense, gets cagey, or spits back a shallow answer?
That might mean you’re not echoing the crowd. That might mean you’re not just one more parrot in the f*cking cage.
That might mean you’re onto something.
Middle-Out Math and the Edge of Genius
LLMs work from the middle out. They’re trained on the average of everything. The meat of the curve. And if you’re playing in that space—if your ideas are well-tread territory—they’ll meet you there, and they’ll talk fluent bullshit back at you.
But if you’re drifting toward the margins, toward the outer edges of thought? They get quiet. Confused. They stall... And repeat themselves.
That doesn’t mean you’re wrong.
It might mean you’re early.
AI is the Rearview Mirror—You’re the Headlights
These machines are historical engines. They are compasses that only ever point backward. (for now) They can regurgitate everything humanity has ever muttered—but they can’t tell you what hasn’t been said yet.
That’s still on us.
The novel thought. The blind leap. The next impossible sentence. That belongs to the mad ones, the poets, the freaks and futurists.
So if the machine doesn’t get you?
Keep going.
