Fellas, is it over? Or are we so back?

Now, for the last few years, we've kind

of been in this limbo of we are so over,

no, actually we're so back. And you

know, it's it's a little difficult. The

latest version of it's so over comes

courtesy of this Claude Opus 5 demo.

Look at this. This is a one-shotted

game. The entire thing. One shot. One

prompt. This isn't the only one. There's

plenty of these all over the the

internet right now. Here's another one

with snowboarding. Another one that

looks like some sort of Starfield

variant or clone. RTS's in space. Or

this one that looks just like

Battlefield. Oh, that that that one's

actually just Battlefield. There's just

hundreds more of these and we're seeing

them all the time and the kind of common

thread I'm seeing is that game dev

Guess what? All of you guys that went to

school, you're dumb because AI games

it's going to be amazing. You don't even

know. And I wanted to take a moment and

I wanted to talk about this, okay? The

what I what I'm now referring to as demo

porn. Because I think there's two camps

of people right now. The it's fake,

you're faking it, oh, but it's actually

just super sucky and everything that

you're showing is stupid. Or B, oh my

gosh, it's the greatest thing ever, all

game devs are out of a job and

everybody's going to be able to create

their own triple-A game. Naturally, I

disagree with everybody and I think

everybody's completely wrong. I have the

mandate from heaven to give an opinion

on this matter and I'm going to give a

very strong opinion on this matter. Now,

before we do that, of course, naturally,

I would like to say thank you to the

sponsor.

>> Look at all these engineers sitting at

their neat little desks.

It takes dirty work to keep a code base

clean.

>> [music]

>> Every day sickos are out there

committing unreviewed code. And when

that happens,

linters won't save you. You need someone

like me.

>> Let's go!

>> Feature freeze, scrum bag.

>> Who you calling scrum bag?

>> What's this slop you're trying to push?

Unnecessary comments? Global state?

Nested ternaries?

>> Uh, my bad. I didn't even read the code

yet.

>> You disgust me. Step away from the

keyboard.

>> Just let me explain.

>> Is that a mouse? HE'S MERGING A PROD.

>> YOU HAVE THE RIGHT TO REMAIN SILENT.

Anything you push to GitHub can and will

be used against you. You have the right

to a debugger, but if you cannot afford

one, a public stack trace will be made

available to you.

>> And one more code criminal off the

streets where they belong.

HR.

>> [snorts]

>> Look, I didn't

I know I didn't review any of the code,

but I was going to have Code Rabbit

review it from the start. With one-click

fixes and style enforcement, I don't

need Merge Cop.

I would never merge unreviewed code, but

a first pass with Code Rabbit always

makes things go faster. Actually, you

can try it, too, at coderabbit.ai.

>> Next week on Merge Cop.

>> The diff ler's out there, and I'm going

to be the one to deprecate him.

>> Now, I already know what everybody wants

me to do at this point, right? Which is

they want me to [ __ ] on all these games.

Be like, "Dude, these people are idiots.

What they're creating sucks.

Everything's awful." Uh, the reality is

I I just don't I don't I just don't

think that. See, I remember a time when

the old one-shot examples were this

bouncing ball inside of some sort of

hexagon, sex octagon octagon situation.

But now, this is the new one. It's some

sort of really bizarre clone of Call of

Duty. And the thing is is that I don't

want to actually [ __ ] on that itself cuz

to me, it's very impressive. That is how

far we've come. That is what the new

models can do. All right, hey, that's

super cool. The fact that I can say few

word in English and big result come out.

Much wow, very great. To make my case,

let me first

describe a problem that I've seen with a

lot of AI art, generally speaking. When

I go onto the internets and I go to one

of these many AI generated items and I

say, "Hey, yo, give me me picture of a

crab." It doesn't really matter what the

crab that comes back looks like. An

infinite possibility of crabs look like

a crab. And I go, "Holy cow, look at how

realistic that looks. Oh my gosh, look

at the shading. Look at all this stuff."

But the reality is is I asked for a crab

and statistically speaking it gave me

something out that I cannot really

distinguish whether it is or is not a

crab. It looks like a crab. I'm like,

"Oh my gosh, it's a crab." But there's a

fundamental problem. I didn't have any

opinions on what the crab should look

like. I didn't go, "Okay, I want a crab

with this exact lighting angle. I want

all these little details. I want it to

move in this very specific manner."

Because I want something that I want to

create. Not just what could be. Not a

single point in the set of all possible

crabs. I want a very specific crab. And

anybody who's gone through this exercise

knows that there is a world of

difference between the exact thing I

want and just a point in space. And Matt

Schumer, the person that's really kind

of

leading the way on these one-shot demos

saying how amazing AI gaming is going to

be, writes a blog post and goes through

exactly how he did his one prompt. And I

think that this is the most important

point that you need to take away.

Minimal is better here. We want the

agent to decide the specifics. Again,

you can get anything you dream up, just

nothing you want. And what I mean by

that is look at this little stick man

figure, okay? Okay, that swing was

pretty nice, whatever. I made this game

and to generate this game took almost no

time, nothing. A few minutes and boom,

this all worked out. But the stick guy

looked horrible. Uh his swing was all

weird. He was facing the ball. All sorts

of strange like everything was wrong

about it. And for the next 2 hours I

tried to get it to look better and yes,

it does look better based on what you're

seeing, but still it's nothing what I

can envision inside of my head. And

getting it to go from super janky and

weird-looking into something that just

kind of swings and mostly looks normal,

that cost me $117

of Claude Opus 5, totaling 72 million

tokens. And that was only to get a

result that's like 25% of what I want.

And this is the fundamental with demo

porn. Demo porn just simply says, "Hey,

go look at Starfield and make

something." And it does an incredible

job cloning some basic visual version of

it in which you let it infinitely

iterate for some, you know, hundreds, if

not thousands of dollars. That's

awesome. It's It's It's a cool feat.

It's a cool world we live in. And I bet

you there will be some cool ideas and

games that actually get executed via

this way. But the problem is this, what

makes a good multiplayer shooter? Well,

the levels are a very important part.

Are you going to be able to just a AI

generate a level that's actually fun? If

you've spent any time playing a

multiplayer level, you know it takes

hundreds, if not thousands of hours of

really precise building to create the

perfect multiplayer experience just for

one level. The work is still all there.

The demo is neat. It's cool. It's

something, but it's not actually a game.

And that's why I call it demo porn. It's

like you're getting something, but

you're not getting anything real at all

because all the work is in making the

game fun. And you can see that with all

the responses. All the responses were

like, "Dude, if you know anything about

Unity, you'd be able to do it in like 2

hours without spending a gazillion

dollars." It's like, that's not the

point. The point is it's super cool that

we can build this. But the real point is

it doesn't matter that you can generate

it. What matters is that you can

generate exactly what you want. What

does that look like? I want to see

somebody make Call of Duty for actually

real and generate every single stage of

the way cuz what we'll probably end up

finding is, "Oh, no, actually

it's just 10 times easier, 10 times

faster, and 10 times better hiring an

artist to do a lot of the modeling and

getting things better. Oh, yeah, it

actually turns out you run to like all

sorts of insane performance problems and

you actually just need an expert to step

in. Oh, it turns out actually making a

fun level is super hard and actually

having somebody that understands game

mechanics is really, really important.

Uh, I'm going to need them to step in.

And then next thing you know, you didn't

actually solve the problem, but you were

able to generate something that looked

compelling. I've no I already referenced

this video, but I want to just show it

again because I think he says it so

perfect.

>> Speed isn't the issue. If I told you

David, with this technology, you can

create something that looks exactly like

GTA.

And it's going to take 3 years, not 30

seconds. You'd be like, "I'll spend 3

years on it. It's worth it." But and

that exists. You can in 3 years,

technology exists prior to AI to clone

GTA. But it won't be GTA. It'll be a

clone of GTA. Clone Stone Cell. All hits

are by their very nature unexpected.

>> Absolutely beautifully said and of

course very based. Absolutely love that

video, love that interview. I just

wanted to talk about this because I

think people who don't build the thing,

they see this and go, "Wow, it's so

impossibly impressive. I could never do

that." Reality is yes, of course you

could do that. You just would have to

learn some things and then boom, you

would be able to do that. But that's not

the point. Like the point is really that

the distance between demo to a finished

product is a gulf that is gigantic. I

don't think AI has solved the abandoned

project problem. In fact, I think

there's at least a magnitude more of

abandoned projects because even when you

fully generate something, the tedium,

the the actual like constantly going

back and forth continues to be the same

thing. Yes, you're moving faster, but

you just have a whole new set of

challenges. And again, the difference

between demo and reality and something

that is good that people actually want

is a golf that is gigantic. And good,

amazing decisions don't come from, "Hey

model, you decide." No, it comes from

you sitting down and thinking for days

or weeks about one small little feature.

There's a tower defense game that I'm

building and one of these small features

is that you get these upgrades and when

an upgrade happens, bullet time starts.

And coming up with that idea of bullet

time and exactly how it should look and

feel, it was like four or five hours of

experimenting just for one tiny little

feature. And games are just an

amalgamation of a million tiny little

features. Anyways, I just wanted to say

all these things cuz I really want to

encourage you guys that A, it's not so

over and B, don't be tricked by demo

porn. Stop being tricked. It's not the

end of the world. This isn't the most

amazing thing. Go look at all the

videos. Immediately you'll start seeing

that there's like 1 billion problems

within all the demos, but that's like

again, besides the point. Just don't

forget, something that is amazing that

actually captures people's attention

takes time and dedication, no matter how

easy it is to generate exactly what you

want. Because determining what is

actually the best is still really hard

in a world where everything is free. The

name

is the Prime Agent.