I Got Bored of Being a Commentator

For a while there, I was good at having opinions about AI.

A lab would ship something. I'd have a take by lunch.

A policy would drop, a model would leak, a benchmark would get gamed - and I'd be right there.

The thread. The framing. The contrarian angle nobody else had quite said yet.

People liked it. The takes did numbers. For a while it felt like a job.

Then one day it hit me that I'd become a commentator.

Not a builder who occasionally commented.

A commentator. A guy with opinions about other people's work.

Somewhere along the way, the reacting had quietly become the thing I did.

And the building had become the thing I talked about doing.

That sat badly with me. It took me a minute to figure out why.

Why commentary is so easy to fall into

Commentary is frictionless. That's the trap.

You can produce an opinion about anything, instantly, at zero cost.

You don't have to be right in any way that gets tested.

You don't ship, so you can't fail.

The feedback is immediate.

Likes, replies, the little hit of getting quoted. It comes in hours, not months.

Building is the opposite. Slow. Uncertain. Mostly invisible until it isn't.

Next to that, commentary is a slot machine that always pays a little.

And the AI cycle is built to reward it.

There's a new thing to react to every single day.

You will never run out of material.

Which means you never have to stop and ask the harder question.

What are you actually for?

I got good at the easy game and mistook it for the real one.

The thing that actually bothered me

Here's what finally got under my skin.

Every take I had about the big labs was, underneath, a complaint.

Too closed. Too expensive. Too willing to flatter you.

Too quick to make something up.

Too comfortable deciding who gets access to the good stuff.

All true, I think.

But a complaint is just a wish with better production values.

I was describing the AI I wanted by listing everything wrong with the AI that existed.

And then publishing it. And feeling like I'd done something.

I hadn't. I'd described a gap.

Describing a gap and filling one are not the same act.

I'd been taking credit for the first while telling myself it was the second.

That's when the boredom set in.

Except it wasn't really boredom. It was closer to embarrassment.

I'm a founder. I've spent my whole adult life building things.

And I'd let myself drift into the one posture where nothing I said had to survive contact with reality.

Founders don't fix things with opinions

The line that snapped me out of it is almost too simple to write down.

If I don't like what the big labs are building, the answer isn't a sharper opinion about it.

It's to build the replacement.

That's the whole difference between a commentator and a founder.

A commentator's position is their take. A founder's position is the thing they shipped.

One costs nothing and proves nothing.

The other costs everything, and proves the only thing that matters - that you actually meant it.

So I stopped writing the take and started building the answer.

I thought AI should be honest instead of flattering.

So I had to build a model that will tell you when it doesn't know.

I thought you should own the thing you talk to, not rent it from someone who can shut it off.

So I had to ship open weights you can download and keep.

I thought it didn't all have to run through someone else's infrastructure.

So I had to build on compute I could actually point to.

That's Vinci. The first model is live now.

It's small. A 4B. The start of a family, not the finish. I'm not going to oversell it.

But it exists. Which is more than any of my takes ever did.

And it does the things I used to just demand other people do.

The lesson

Opinions are cheap.

I don't mean that as a slogan. I mean it almost literally.

The market price of a take is about zero, because anyone can make one and none of them are backed by anything.

The only opinion that carries weight is the one you were willing to build.

Because that's the only one that cost you something to hold.

Commentary is a comfortable place to hide.

It looks like contribution.

It feels like influence.

It gives you the posture of being in the arena without the risk of being in it.

And the longer you stay there, the more you start to believe that having the right opinion is the same as doing the right thing.

It isn't.

Building is the only opinion that counts.

I'm not swearing off ever saying anything again.

I'll still think out loud - that's part of how I work.

But the center of gravity has moved.

The take is the byproduct now, not the product.

The product is the thing I ship.

What I'm watching for next

The pull back toward commentary, mostly.

It's strong, and it's seductive, because it pays faster than building does.

I can already feel it.

A lab will ship something next week, I'll have a take, the take will do numbers, and the slot machine will offer me a comfortable, hollow little win.

I want to catch myself when that happens.

And ask the founder's question instead of the commentator's.

Not 'what's my angle on this.'

But 'if I actually disagree, what am I going to build about it.'

Vinci is my answer to that question for now.

It's a start. And I'm glad to be back to building it - instead of just talking about whether someone should.

— George