It’s 2 AM, and the feed is doing its thing. Your college roommate just posted about their third startup exit, someone from high school is in Bali “living their truth,” your cousin’s page is wall-to-wall pumpkin patch. Everyone’s life reads like a clean sentence. Started here, which led to here, which led to now.
Your life does not read like that. Marketing analyst, failed food blogger, product manager, six months of pottery instructor. None of it connects, so you close the app, stare at the ceiling, and wonder if maybe you’re reading it wrong.
Two ways to make meaning
Borrow something from AI for a second.
Autoregression
LLMs like ChatGPT and Claude write one token at a time. A token is roughly a word. Every token depends on the whole history that came before it, all the way back to the first one.
These models don’t know the capital of France is Paris, because they’ve seen “The capital of France is” followed by “Paris” enough times that they predict it. It’s pattern matching, all the way down.
It’s how you can sometimes guess the next lyric of a song you’ve never heard, because your brain is running the same prediction over what comes next given everything so far.
Write “The cat sat on the…” and the next word is probably mat, chair, or floor. Each word you commit narrows what can come after it.
Diffusion
Image models like Midjourney and DALL-E work differently. They start with pure static and refine the whole canvas at once, thousands of passes, each one asking what this would look like with slightly less chaos.
Think of streaming a video on bad Wi-Fi. First it’s smeared, then at 480p you can tell there are people, then at 720p you know who. Diffusion does the same thing in iterations. At step 50 you can make out shapes, at step 70 a person, and at step 90 the model can still surprise you, a sunset clarifying into a sunrise, a cat turning out to be a fox.
Every pixel adjusts based on every other pixel, every pass.
Language has to be autoregressive, because a story needs causation. This happened, so that happened. Without the chain, it’s just a list.
An image works the other way around. You take in a painting all at once.
So when you think about your own life, which one does it actually look like?
Your life is the image
Your relationships, skills, memories, and hopes all happen at the same time. They sit inside each other. Your career bleeds into your friendships, which bleed into your taste, which bleeds back into your career.
The breakup three years ago is still a pixel in your current ambition. The book you read last month is quietly rewriting how you remember being twelve. Your friendship with Sarah is shaped by what you’ve shared, what you’re doing now, and what you imagine doing later, all at once.
This is diffusion. The whole picture moving together, slowly clarifying, the iterations called days and years.
Everyone lives this way, including the LinkedIn superstar with the perfect arc. What changes from person to person is what happens when they sit down to describe it.
Tell me about yourself
Society runs on autoregression. Resumes, profiles, party introductions, “tell me about yourself” all demand a sentence. We all face the same task of taking a diffusion reality and translating it into a story that pretends one thing led to the next.
Some people’s pixels happen to land in shapes that look like sentences when you squint.
Engineering degree → tech job → senior engineer → team lead → VP.
The arrow is doing the work. It implies the degree caused the job, the job caused the leadership. The reality probably included a few near-quits, a serious alpaca-breeding phase, and a lot of luck. None of that survives the arrow.
These people are living the same kind of life you are. They’ve just learned to pull a cleaner slice out of it, and they pull different slices for different audiences. LinkedIn gets the victory lap, Instagram gets the grateful coffee shot, Twitter gets the contrarian take. Same diffusion process, three autoregressive outputs.
Your image is rendering too, but the shapes haven’t landed yet in the kind of order that makes for a clean “and then, and then.”
Mid-render
When someone asks what you do, you’re being asked to describe a half-formed photograph. You can see shapes. The marketing job, the pottery, the food blog are pulling toward something, but you can’t name it yet.
So you do one of two things. You make up a story that isn’t really there, something like “clay is a lot like user experience.” Or you tell the truth, and the other person’s eyes glaze over halfway through.
Meanwhile you’re comparing your unfinished image to other people’s finished captions. Rough draft against published novel, forgetting they had twenty drafts you never saw.
Believing it too early
The damage starts when you believe the story you forced.
“I can’t stick with anything.” “I’m a failure who keeps changing directions.” “I’m lost.”
That’s what happens when you try to autoregressively narrate static, because the only stories that fit static are stories of failure.
You can stop, because the image isn’t done, and any story you tell about it right now is fiction either way.
Iteration 95
A diffusion life clarifies the way a diffusion image does, gradually, with the last passes still able to change everything.
You might spend years thinking your pottery phase was about creativity, and only later realize it was about patience. The food blog you filed under “failure” might turn out to be where you learned how to talk to an audience. The marketing job might be where you found that you like helping makers tell their own stories.
Or it might have just been a paycheck. Some pixels stay pixels.
At iteration 95, the picture is clear enough to describe. Pottery taught patience, which made you good at user research, which made you better at food blogging, which gave you the angle the marketing job needed. That’s the moment the story writes itself, and not before.
This is why memoirs work and life advice often doesn’t. The memoirist is describing a mostly-finished image, while the advice-giver is trying to help you generate tokens before your pixels have resolved, usually from an image that clarified in a way yours won’t.
In the meantime
A few things follow.
Your daily experience is diffusion, always. Every choice and every random Tuesday is adjusting every other part of the picture. You won’t know which Tuesdays mattered until much later.
You still have to generate autoregressive stories for the world, the kind that interviews, profiles, and dinner parties need. Just don’t believe them too hard, your bio is fan fiction you’re writing about yourself.
Use different timescales. Stories can update monthly, the image needs years, and you cannot judge a render at iteration 50.
And find people who can sit with you while the image is still forming, without needing you to hand them a sentence.
2 AM
Back to the ceiling, back to the feed. Your roommate, the Bali friend, the cousin are all showing you one clean thread, pulled out of their static, the version that fits a dinner party, the version that fits a status update.
What they leave out is what you leave out, the 2 AM scrolling, the failed projects, the confusion about what any of this is for.
Nobody posts “Still trying to figure out what this all means” when they can post “Excited to announce.” Nobody’s bio reads “Just winging it with better lighting.”
Your static hasn’t landed in a shape yet, because you’re earlier in the render than they are.
Close the app. The pottery, the marketing, and the food blog are all pixels in an image that isn’t done rendering. The shape will become clearer with more iterations, and some of those iterations are the boring Tuesdays you’re already living through.