Today I got to try out Opus 46 and Chad

Chippity 5.3. Okay, these are the latest

and greatest state-of-the-art models. It

is kind of funny because at 10:45 a.m.

today, Opus 46 launches, but then at

11:12 a.m. Jippidity 5.3 Codeex

launches. I there's something about it.

I'm not going to lie. I think Sam is

feeling a little bit jealous. Okay, he's

a little upset at Daario for making

those commercials. Not very happy for

him getting made fun of. Yet, here we

are. Anywh who, I built the same

application in both of them. As you can

see this beautiful application right

here and this beautiful application

right here. How I did it is I asked it

to make a JSX transformer that could

take in JSX and produce out a 60 frames

pers terminal application that's

rendered using bun and I want the

transformer to be written in Rust for

JSX to JavaScript. I wanted to have hot

module reloading. It involved making a

bunch of prompts to both. But I did the

same initial seed prompt, let it do plan

mode. Jippy asked a couple extra

questions, got those done. Opus only

asked me one question, got those. And

then all the follow-up answers were all

answered in the exact same order doing

the same thing. So as close of a, you

know, sanitized test as one could make.

Also, what what is this? What what what

did what did Claude draw here? What is

this? I don't I don't know what to do

with this. Ridiculous. So how did it

actually do? Well, first I want to point

out that chat jippity 53 it actually did

make some JSX. It actually has on the-ly

compiling. It actually had quite a few

things. Never got hot module reloading

working. But I can come in here and I

can simply adjust this thing right here.

Resave it. Just rerun it. And you can

see the thing moves. Actually pretty

impressive. I am pretty happy about it.

I even asked Opus about it. Like, "Hey

Opus, how did Chad ship 835 do?" And

it's just like, "Well, it kind of

cheated on the syntax. It was doing more

of a DSL than anything else." Also, I

just want to let you know that it did it

was kind of it was kind of creative

approach because it did sidestep the

entire JavaScript ecosystem with JSX

tooling. And you know what I said? I

said, "Hey, that's smart." Okay,

JavaScript ecosystem for tooling. That's

what I have to say about it. On the

other hand, this is Opus'. Okay, first

off, art not that good. Okay, Squid,

Crab, and what appears to be LA. Okay,

this is LA. You know, the old dongle, I

guess, represents LA. Well, its code's a

little bit different. First off, you'll

notice something right here. Uh, I hate

to tell you this. That ain't LA. It just

cheated, which I'm not going to lie to

you, not very happy. I'm not very happy

at all. It just cheated. It just uses

functions right here. This is not uh

this is not the compiled code. This is

just the actual running code. But I will

say on the other hand, it did get hot

module reloading actually working. So

hey, little bit better for Opus on some

of the things. But the nonJSX thing, not

very happy about it. Also something kind

of interesting. It took about 2,000

lines of of of JavaScript to make this

happen, whereas it's only about a,000

lines of JavaScript for chat GPT. Also,

the chat GPT1 has an actually working

JSX parser also written with only 520

lines of Rust code. Whereas the compiler

for Opus, which I mean I don't know

what's going on here. It's not actually

running. It's not actually doing JSX is

1,300 lines of code. So, is that a good

measurement? I can't really tell. One

for one features. It's really really

hard. But I can say there was live

actual JSX compiled in one of them in

fewer lines of code. That's a pretty

good measurement for me. Also, just

generally speaking, I tend to prefer

ChatJippid's code output comparatively

to Claude. I can't tell you why. I just

seem to like its styling just a little

bit better. I think it just tends to

organize things a little bit better. Its

functions aren't maybe as as egregious

as Opus. Opus just is like, "Dude, I'm

solving. I'm solving. I'm solving. Next

function, I'm solving." Like, it just

doesn't feel as good ever looking at the

code. So, honestly, I I feel like Chad

Chippity kind of won here. Now, at the

end of the day, does it actually matter

which which model you use? I actually I

don't think the state-of-the-art models

at this point really matter. Long as you

you hold it right, it'll just work. I

think at the end of the day, if you're

going to use a model, they're all really

really decent, right? They can all go

pretty far producing code. And at the

end of the day, it's on to you to making

good code. And it's it's like the same

old thing. It nothing has really

changed. The people that weren't

producing good code just produce bad

code faster. the people that were

producing good code, they produce good

code and probably not at an entirely

changed rate, like a decent small like

maybe some small chunk increase because

producing good code is actually really

hard. So, is it going to matter if

you're on 45 versus 46 if you're still

stuck on like 51 codeex versus 53

codeex? I don't think it's going to

really matter. I think at the end of the

day, if you know what you're doing and

you practice, you get in those reps, you

get the time in the saddle, and you work

really, really hard at your craft, I

think you're going to be happy no matter

what model you're using, even if it's

unbreakable Kimmy K. And I think that's

the real takeaway here. And I also think

that that's why you see these model

providers competing so hard because when

one does an improvement, the other one

needs to because there's not really

anything holding people there, right?

It's just like, okay, I like this model.

I like this model. No, I like that

model. Honestly, I think one of the big

things is that people have a good

experience with one model and then

they're like, "Oh, this is the best

one." And then you just kind of stick

with it for some sort of weird emotional

appeal for some amount of time and then

it does something. You're like, "I don't

like it anymore. I'm going to try this

other one." And then it does something

good. You're like, "Nah, this is my new

favorite one." I just wanted to throw

that out there because I think you're

going to get hit with a bunch of

benchmarks over the next few days. Not

even sure any of them really matter at

this point cuz let's just face it when

you're producing 10,000 nay 15,000 lines

of code a day. Does it really matter? I

just don't I'm just not sure it matters

at all. And a lot of that's because I

just have this theory. Okay, now we're

going to get into some weird theory

time. Okay, I think that the average

programmer is of quality X. Now, here's

the weird part about the quality of X. I

am not really sure if it's between 0 to

one. I think it might be between

negative one to one where some engineers

and you know this you've worked with

some engineers when they build things

afterwards it actually takes more work

and more time to figure out how to undo

some of the things that have been done

and I kind of feel like AI it's more of

a multiplier right it's not an addition

it's not like plus five you're not just

that much better it doesn't take

somebody that's a negative0.5 and turns

them into a 4.5 whatever these numbers

mean developer no it's more like a mult

multiplier. So if you work with somebody

that say is a negative0.1, you know,

like they deliver features, but there's

a bit of kind of tech debt that kind of

comes associated with these features,

the only thing I think you're getting is

you're going to be transferring them

from a 0.1 to a one. They're just going

to make those questionable features just

10 times faster. That's kind of like my

grand unifying theory of programming as

it is along with kind of the impacts of

AI. And I also think I know there's

going to be another crazy one. I also

think that the impact of AI is

proportional to how much you're on this

side of the equation versus this side of

the equation. Like there are some things

that man it makes life easy. Like yo, go

run this integration test and then I

want you to figure out what's breaking

about it and then I want you just to

give me a report of why you think

something breaks. I'm going to come back

to you in 30 minutes and I'll see if you

figured out some stuff and kind of

follow your leads. Oh my gosh. Saves me

literally hours a day. Like that that

actually is an incredible feature of AI.

Okay, that's it. That's all I have to

say. Okay, bye-bye. Take care. The name

The name is I don't really The name is I

don't I don't really care. A Jen. Hey,

is that HTTP? Get that out of here.

That's not how we order coffee. We order

coffee via ssh terminal.shop. Yeah. You

want a real experience? You want real

coffee? You want awesome subscriptions

so you never have to remember again? Oh,

you want exclusive blends with exclusive

coffee and exclusive content? Then check

out cron. You don't know what SSH is?

>> Well, maybe the coffee is not for you.

Living the dream.