It's almost 2025, and if there's one thing that's consistent among developers, it's creating new languages that are kind of like 'C, but...'.  You know, 'C, but easy,' 'C, but modern,' or 'C, but for furries.'  But this time, we have a language

designed to be used with AI.  That's right, it's going to be 'C, but for AI and furries.' Obviously, AI is going to take all of our jobs the moment that product managers can accurately describe what they want, which means our jobs are effectively safe at infinitum.  But there is

room for us to be able to take advantage of the AI, you know, the classic AI experience that we all have where we tell the AI that it's a mother,  we're proud of it, we glaze it really hard, or threaten to commit seppuku if it doesn't give you

back JSON. And yes, that still is true when it comes to Anthropic.  Right now, I cannot get it to properly format a JSON.  Anyways, this new language is called Mir, and it's kind of unique, right? It has this basic grammar that proposes signatures, which is kind of what the

function name and shape should be, and then examples. So kind of like test-driven development, where you don't write tests, you just write like the blackbox version of it. That's right, that means it's an AI-powered functional programming experience in JavaScript, of course.

It produces JavaScript. I mean, what other language would it have produced? It only produces JavaScript. That means you get all the benefit of avoiding writing code plus you get all the agony of writing tests. It's late-stage TDD, really. That's what it is. We've officially hit

late stage AI powered test driven development and if a VC ends up seeing this video, pretty sure Mir's in store for about a $10 million funding because let's let's just be real here AI that's all you got to say that's actually all

you have to say to a VC and you you get millions of dollars. I mean look at friend.com. Anyways, I played with Mir for about 2 hours in hopes of making something, really anything, just like can I even create the simplest function? But as I started playing with it, I realized

several things: one, it does appear that an AI has written this language because the specification is actually insane. Look at this: literals, literals is one literal and an optional zero or more literals. What syntax is that? I don't know, but you eventually kind of figure it out. You go

on over to literal and you look at the list type, you'll notice that it's in brackets: literal. Okay, so what does that mean? That means you can only produce a list of one item. It's just an inconvenient variable. Like oh my gosh,

what is this specification? And when of course when you look into the types, you'll see right here that you can produce a list with a type or a dict with a type. What, what, why, why are there malformed dicts? I never thought I would ever see a malformed

dict, but here I am, 2024, almost 2025, and I can't stop singing: what is a one-typed dict? What is a one-typed [Music] dict? Is that like supposed to be a set? That's not a dict, dog! What is, what is

that this specification is a lot like kind of buying off-brand cereal, you know it has all the proper shapes and everything, but the moment you actually try it out it's not the same, it hits differently, it does not feel good. So of course I did the only thing that I could

possibly do, a fix the list boom, also turns out you can only have positive integers, so fix the integers and look at that, what is that? Is that a red X? Yeah, that's right, there's a red X in a parser. How do you build the tokenizer

and then you're using a red, oh no, yo dog, I heard you like parsers so I put a parser in your parser. This project clearly has a lot of AI influence because the types and the setup is just, it just feels all crazy, but you know what? We got it fixed, I didn't

fix the dicks though, I don't think anything can fix that dick. That is one malformed dick, so I left that aside and I figured okay, let's solve a problem using lists as my data structure. I know convenience, developer experience, all wrapped up into one data structure, lists.

All right, so the problem I decided to do was AABB collision with a point. All I want is an axis-aligned bounding box and a point. Also, thank you for the sub, appreciate that. Hey Zippy, thanks for that offline sub, Zippy. All I want is an axis-aligned bounding

box and a point. I'm simply going to have something that takes in a list of height, width, and its X and Y position, and then it's something else that's just a list of XY position. And now I'm just going to have to specify a bunch of examples and

hope that it produces out the code that I wanted to produce so here right here is my first attempt and here's the code that it produces now just one quick look at it we'll see that it produced just exceptionally incorrect code and I know right now I can hear them I can hear

them charging all the way in starting to yell getting all frothy at the mouth the AI proed Max Melissa like well actually you didn't give you didn't give enough prompting okay that's the problem is you didn't give enough examples your examples

weren't good enough I don't think you actually know what you're doing see the thing is if you just understood prompting you'd be able to prompt better and clearly you don't understand prompting of course I don't it's brand new language that I had to fix

myself so I'm going to give more examples and obviously a second clear problem with this approach is that there's no like check boxes where I can say how much and by what extent I wish to either threaten self harm harm to the mother or glaze the AI so obviously

probably not going to get it correct and does it even tell the AI not to hallucinate another really valuable strategy I don't even think it does anyways gave a whole bunch more examples and got a new piece of code which in fact still does not work if you couldn't

understand why this code was wrong look at it input zero let's call that X and input one let's call that Y if input X is greater than any element within that element list and input two is less than or equal to the element adjacent to it

It's considered containing like that's crazy, like how did they get that? How did it get this answer at this point? I'm two hours into fixing bugs and producing absolutely nothing, so I just tried GPT. Gave it a quick little English sentence.

Gave it the structure of my code, boom, one shot of the answers right out. You know this project's kind of like cute, it's funny. Hey, it's a great way to learn how to write a parser, so if you, whoever wrote this, actually ended up writing out

the parser by hand and doing everything, it's a great experience really to learn and to become a better programmer. So hey, hats off to you, but at the end of the day, this always highlights the same thing for me, which is why do men avoid learning to code? What is this like, why

are you doing this? Just learn the code, just learn how to just test it, a points in a box, learn to write a return statement with a couple conditions. It's like not hard. You can go so much further if you don't have to rely on something

else to think for you. I just don't understand it, and here, hear me out here. If these AIs were really that great, they're so great that they actually were AGI, they're able to learn, they're able to grow better as Sam Altman once said.

We're only a couple thousand days away from it. That's like 10 years, anyways. Also, that's a crazy way to say that. Who, who describes time windows in days like, 'Oh yeah, it's just, it's just like 3600 days away, 10 years.' You can

just say 10 years and no one's upset at you Samuel it's strange but anyways but even if these companies do produce something that actually can really produce good code you just got to understand something why would a company sell a product that can build anything why

wouldn't they just harbor those trade secrets and build everything right like why would you sell the infinite intelligence tool think about it for a couple it makes no sense like you would be able to compete with every product and start company after company after

company after company over and over again having super AI would be the most valuable thing in the universe so why ever would they just simply sell it they wouldn't because they'd be able to make everything at the speed of like light it'd be crazy let's just say that

the AI companies do fumble the bag and accidentally sell the greatest product ever that can produce anything in a snap of the fingers with just a basic few text prompts let's say that does happen you having hard skills is going to be absolutely incredible right like

learning how to be technical and talented and solve problems large on large scales will only help you it will only make you better it will only make you more valuable and when that inevitable day comes whether it's in 10 years or a thousand years that AI can

just do so much your technical skills will still be valuable stop making the trade it's not a trade you don't have to make the trade like you really don't you can just like learn and have fun and build stuff and become better at your craft the

Programming Is Cooked

This is a place for all the things that are awesome on stream. 

ThePrimeTime

Introduction to Mir, an AI-powered programming language

Mir is a new programming language designed for use with AI, described as 'C, but for AI and furries'.

The language's design allows developers to leverage AI capabilities.

Mir uses a unique grammar with proposed signatures and examples, similar to test-driven development.

It's an AI-powered functional programming experience that produces JavaScript.

The speaker speculates about potential funding for Mir based on its AI capabilities.

Experiences and observations while using Mir

The speaker spent two hours trying to create a simple function in Mir.

The language's specification is considered unusual and complex, with examples of confusing syntax.

Issues with list and dictionary types are noted, including malformed dictionaries.

The speaker compares the specification to off-brand cereal, highlighting inconsistencies.

The speaker fixes issues with lists and integers, but encounters a parser error.

The speaker attempts to solve an AABB collision problem using lists as the data structure.

The initial attempt produces incorrect code.

The speaker adds more examples, but the generated code remains incorrect.

After further attempts, the speaker uses GPT to solve the problem easily.

Reflection on the Mir experience and the role of AI in programming

The speaker reflects on the learning experience gained from working with Mir's parser.

The speaker questions the need for AI-assisted programming, suggesting that learning to code directly is more efficient.

The speaker speculates on the future of AI and its potential impact on the job market.

The speaker questions why AI companies would sell a product capable of creating anything, suggesting it's more beneficial to keep it internal.

The speaker emphasizes the continued value of technical skills, even with advanced AI.