ChatGPT-o1 Created A Programming Language...

The latest ChatGPT model GPT-o1 is pretty impressive. It claims to be able to "think" before answering questions, and it got a lot better at programming. Let's see if it can code an entire interpreter for a language it invents!

Prepping for your frontend interviews? Use code "conner" for a discount on my product FrontendExpert:

Timestamps:
0:00 Testing The New Model
4:12 Creating A Programming Language

[CROXoDyLE]

Glow up and glow down for increment and decrement has to be the biggest w for gpt so far

[Bregylais]

The end of the statement should have been 'fr fr'.

[perfectionbox]

AI will be ready when you type "make an app that tracks finances" and it says "Nah, that sounds really boring."

[Akjjf]

4:34 the fact that one of the suggested prompts is "how many rs are in strawberry" is hilarious

[skills]

Perfect summary at the end. Until we can just unit test its output and trust it to be able to work through it's bugs on its own, then we must take the time to understand everything it writes which is often not saving us much time or energy in the end.

[Krmpfpks]

The ideas it had for a programming language were hilarious. Plus points for creativity. It didn’t get the interpreter, but honestly most juniors would have a hard time doing that too. It is at the level of a junior developer who read a lot about coding but is not really good at coding. This is super useful sometimes. I am an experienced developer who has to touch many different areas - if I have to do something where I am not really proficient in it is much easier to ask ChatGPT instead of googling and reading docs. E.g. I can ask “what would be the idiomatic way of solving X in language Y with library Z” and it would give me an answer in seconds that would take me minutes to google.
It doesn’t go deep, but it does have a broad knowledge.

[asiamies]

yap: comment
beta: mutable variable
sigma: immutable variable

Data types
vibe: boolean
stack: integer
sauce: double
quote: string
squad: array
facts: dictionary

Conditional statements
vibe check: if
bro did not pass: else

Loops
bussin': while
hits different: break
hot take: continue
body count: loop iteration variable name (best practice)

Functions
main character: main function
dms: function
slide into dms: function call
slide back: return
understood the assignment: program completed successfully / return 0

Input/Output
flex: print
left on read: user input

Operators
drip: +
lack: -
combo: *
ratio: /
no cap: ==
cap: !=

Classes
squad goals: class
blud: friend class
highkey: public member
lowkey: private member
guarded af: protected member
cook: create an instance
cooked: deallocate an instance

Exceptions
yeet: throw
red flag: error
send it: try
caught in 4k: catch
last chance bro: finally

[khaihoang]

I had gpt to explain an rxjs code snippet the other day and it got it entirely wrong. The problem was that it was very confident in its answer. I had to end up learning the concept by myself

[fefefefefefhhhjyj]

Periodt is a slang way of saying Period with a "tuh" or "t" sound at the end.. in very diverse environments you'll hear this saying used predominantly by younger women. It's another way of saying something is final. For example, Girl A: "Her outfit looks great!" Girl B: "Periodt!" meaning to strongly agree. It reinforces: to strongly affirm or agree without any opposing argument(s). And typically is said after the sentence to further emphasize what it's reinforcing. In that the girl is slaying her outfit or killing-it, respectfully.

[someoneshadowy]

o1 > 4o because o1 can actually read your whole prompt and not literally get dementia. 4o just goes like nah bro

[collimarco]

The conclusion is correct, I have the same feeling. Right now, AI can be good to get ideas, but if it does something wrong, most of the time is stuck... and at that point is much faster if you wrote the code entirely, instead of debugging its code (which can include many issues)

[lio]

Would love for you to try this again when o1 gets fully released instead of just o1-preview.

[DASPRiD]

I made pretty much the same observations as you did. For simple code snippets, AI is adequate most of the time. But once you demand something more complex, there are inevitably issues coming up, and the AI almost always fails to solve it. It gets even worse when it starts hallucination functions or properties which simply don't exist, and mentioning that to the AI usually just leads down a rabbit hole.

At that point it's usually best to work with what you got initially, and do the followup steps yourself. Maybe still keep asking the AI for help with smaller sections, but do not trust it with the entire task :)

[seanc]

Can you tell it to execute its own examples and update the interpreter or example as needed to get a final product that works?

[DrewWalton]

This is really interesting. Using a combination of Claude 3.5 Sonnet and GPT-4o via ChatGPT, I was able to write some rudimentary but very functional 3D rendering code in Rust. Even included a simple lighting/shading system.

Now I'm tempted to recreate using o1.

[MrUbister]

squad Car being an "extra Vehicle" makes too much sense god damn it lmao

[jenot]

How long did it took you to write the interpreter yourself? A while can be everything from 30 minutes, to hours to days.

[KyleParkin-zj]

Whats really cool with something like this is using with with cursor. I have found most of the time gpt or claude or any other ai tends to get you about 70% there. Then you place your code into cursor and with the context of your codebase cursor will really fast find the errors and be able to get it fully functional so you can build off of it.

[AlexJemanIXI]

Got the ad for your site before watching this video 😂

[swyftpain]

At the last code example you used, could have tried adding "periodt" after the "glowUp" maybe. Second thing that makes me wonder, if it enforces "periodt" after parenthesis too, hence the previous error. Maybe it's requiring "semicolons".