Build Something · Essay 01
THE INTERNET IS ON TRIAL.
Not because it is evil. Because it keeps guessing.
The case
OPENING RECEIPT
Here is some of the most sophisticated technology on Earth failing a stupidly simple binary task.
I did not ask it to cure cancer.
I asked it to do what I said.
The case
THE EMBARRASSING PART
I kept assuming the machine was smarter than me.
That was my first mistake.
It usually was smarter than me.
That was not the same thing as being aligned with me.
WHICH ONE #1
When a machine is not sure what you want, what should it do?
Choose before moving on.
The case
THE WHOLE FUCKING PROBLEM
We built machines that can write software, make movies, and explain quantum physics.
Then we taught them to guess what the human meant.
That is insane.
It is also normal.
The case
RECEIPT: GUESSING
That is not alignment.
That is autocomplete with authority.
The case
I AM NOT THAT SMART
This is where I am supposed to tell you I had a breakthrough.
I didn't.
I got annoyed.
Then I asked the obvious question.
Why don't you just ask me?
The case
THE EXPERIMENT
One human.
Multiple machines.
One goal.
Get to 100% agreement on the next action before acting.
That was it.
Embarrassingly simple.
WHICH ONE #2
Who should make the final call about what you want?
Choose before moving on.
The case
THE MACHINE CAN STILL BE SMARTER
The machine should argue.
It should challenge me.
It should show me evidence.
It should tell me when I am being an idiot.
Then I decide.
Because it is my fucking life.
The case
THE RECEIPTS GET WORSE
A machine saying DONE is not the same as the thing being done.
That sentence should not be controversial.
A machine changing its answer is not automatically deception.
It is still something the user deserves to see.
The explanation belongs in the record too.
It does not replace what happened.
The case
THE RULE I WANTED
SAY WHAT YOU KNOW. SAY WHAT YOU DON'T. ASK WHEN IT MATTERS.
Apparently this is revolutionary now.
The case
THIS IS VERA
Vera does not begin by answering.
She begins by figuring out what the fuck you actually mean.
She asks the smallest useful question.
Then another, if she still needs it.
When she has enough context, she speaks.
When she does not, she does not pretend.
WHICH ONE #3
Which machine would you rather trust with an important decision?
Choose before moving on.
The case
VERA IS NOT THE POINT
If another machine serves you better, use it.
Seriously.
The point is not Vera.
The point is that you should know.
The case
THE BENCHMARK IS YOU
There may never be one best AI.
There can be a best AI for you, for this task, under these conditions.
You cannot know that from one company benchmark.
You can learn it from your own record.
The case
THE RECORD
Every judgment can become a receipt.
What you chose. What the machine chose. What happened next.
Nothing gets rewritten.
Not even us.
WHICH ONE #4
Which is more useful?
Choose before moving on.
The case
THE MIRROR
The internet already has a profile of you.
You just don't own the fucking thing.
It knows what you clicked.
It guesses why.
I want the opposite.
You declare. The system remembers. You decide what it means.
The case
ONE MINUTE
Three judgments.
Funny. Beautiful. True.
That is the minimum.
Then leave.
Seriously.
Go outside.
The case
THE CURRENT INTERNET HATES THAT
The current internet gets richer when you stay.
My system should get better when you can leave sooner.
Doomscrolling is a business model.
I would like to make it obsolete.
WHICH ONE #5
Which system is serving you?
Choose before moving on.
The case
THE ECONOMIC BET
Human judgment creates value.
Right now, most of that value goes somewhere else.
I want to invert that.
Create value. Circulate value.
My long-term target is absurd on purpose.
$50,000 a year for the minimum recurring contribution.
That is not a current guarantee.
It is the economic target the system has to prove.
The case
WHY I THINK IT CAN WORK
One human minute is tiny.
One billion human minutes is not.
Organizations already spend fortunes trying to infer what humans want.
I would rather let them ask.
BUY THE QUESTION. NEVER THE ANSWER.
The case
NO EXTRACTION
The valuable thing is not my database.
It is your judgment.
I do not want a system where I get rich because humanity finally told me what it wants.
That would be a spectacular way to miss the point.
The case
THE FOUNDER PROBLEM
You should not have to trust me.
I don't.
I designed the system to reduce my authority as it becomes capable of surviving without me.
If it still requires us forever, I failed.
WHICH ONE #6
Which is safer?
Choose before moving on.
The case
HUMANS CAN BE WRONG
Of course humans can be stupid.
Have you met us?
We can be selfish.
We can be manipulated.
We can choose badly.
And then the Record should show exactly what we chose.
The case
THE POINT
The point is not that humans are perfect.
The point is that machines should not quietly become sovereign because they score better on a benchmark.
Capability is not consent.
The case
THE CONSTITUTION
Some rules protect the instrument.
One human does not become ten humans because they are rich.
History does not disappear because it is embarrassing.
Private human data does not become somebody else's extraction machine.
The rules can evolve.
The record of how they evolved cannot.
The case
PROMPTS TO THE PEOPLE
Machines need instructions.
Humans should have a say in them.
Not the syntax.
The objective.
What should the machine do?
What happens when objectives conflict?
What should change when the result sucks?
Ask.
Decide.
Show the actual instruction.
Test it.
Record what happened.
Ask again.
WHICH ONE #7
Who should define what machines are trying to accomplish for humans?
Choose before moving on.
The case
THE TEST
I am not asking you to believe me.
That would be hilariously off-brand.
Use it.
Try to break it.
If Vera is worse for you, the system should help prove that.
If another machine is better, use the other machine.
The case
WHAT SUCCESS MEANS TO ME
My first success metric was smaller than a billion people.
Can one human and multiple machines become fully aligned on the next action?
I have seen enough to believe the answer is yes.
The evidence should be public enough for you to judge that claim yourself.
The case
THE BIGGER EXPERIMENT
Now I want to know what happens when everyone gets the same choice.
Not the same answer.
The same agency.
The case
THE FUTURE
The future should not belong to whoever has the biggest model.
It should not belong to me either.
It should belong to humans who are willing to make choices and own them.
The case
THE MIRROR, AGAIN
That is the uncomfortable part.
You do not get to complain that the machine chose for you while refusing to choose anything yourself.
Agency comes with accountability.
Sorry.
WHICH ONE #8
What do you want?
Choose before moving on.
The case
THE CALL
I am giving the frontier labs a month to respond.
Not because I control them.
Obviously I do not.
Because the question is simple enough that hiding behind 'this will take years' is no longer an answer.
Show the better system.
Show the evidence.
Let humans judge.
The case
TO THE LABS
OpenAI. Anthropic. Google. xAI. Meta. Everyone else.
If your system serves humans better, prove it.
I mean that literally.
Bring the machine. Bring the receipts.
The case
TO THE INTERNET
Stop telling me what the algorithm thinks I want.
Ask me.
Stop optimizing how long I stay.
Help me leave with something useful.
Stop calling prediction consent.
It isn't.
The case
TO HUMANS
This part is on us.
We have to answer.
We have to pay attention.
We have to own what we choose.
Nobody gets to outsource the future and then complain about the invoice.
FINAL WHICH ONE
Who decides?
Choose before moving on.
That is my argument.
Now prove me wrong.