I’ll admit it — the first time I asked GPT to write a Python function, I felt like I’d just hired the world’s fastest, most polite junior developer.
- “Write me a Flask API?” Done.
- “Make it work with async?” Done.
- “Can you also add some unit tests?” Already finished before you blink.
Then I got cocky.
- “Hey GPT, write this in Assembly.”
What I got back looked like it had been through three rounds of telephone, translated into Klingon, and then copy-pasted from a blog post last updated in 1998.
Why the Drastic Personality Shift?
1. Training Data Bias
GPT has been binging Python and JavaScript code for years. GitHub, Stack Overflow, tutorials — all loaded into its brain. Assembly and Rust? That’s like a dusty shelf in the library nobody visits unless their professor forces them.
2. Forgiveness Levels
High-level languages are chill. You can hand Python something slightly broken, and it’ll still run. Assembly? One wrong instruction and the CPU basically says, “Nope. Not today.” Rust? The compiler’s motto is “Prove to me you deserve my trust.”
3. Error Handling
When GPT writes bad JavaScript, you fix it in 5 minutes:
javascriptCopyEdit// GPT output
function greet(name) {
return "Hello, " + name;
}
console.log(greet("Alice"));
// ✅ Works perfectly.
When it writes bad Rust:
rustCopyEdit// GPT output
fn greet(name: &str) -> String {
format!("Hello, {}", name)
}
fn main() {
let mut name = String::from("Alice");
let greeting = greet(&mut name); // ❌ borrow checker meltdown
println!("{}", greeting);
}
Compiler: “You can’t borrow name as mutable because you’re already borrowing it as immutable, and also because Mercury is in retrograde.”
When it writes Assembly… brace yourself:
asmCopyEdit; GPT output
section .data
msg db 'Hello, world',0
section .text
global _start
_start:
mov eax, 4
mov ebx, 1
mov ecx, msg
int 0x80 ; ❌ missing length, outdated syscall, probably for Linux 2.4
I’m 90% sure this code last worked back when flip phones were cool.
4. Library Fluency
Ask GPT for a React component:
javascriptCopyEditfunction App() {
return <h1>Hello, world!</h1>;
}
export default App;
Ask it for ARM Assembly:
asmCopyEdit; GPT output
MOV something, somewhere ; 🤷 "I think this is how it works?"
Where GPT Works Best
- JavaScript & Python → Prototypes, automation, even production-ready code (with human review).
- Assembly & Rust → Pseudocode, vague scaffolding, or something to laugh at before rewriting it from scratch.
Takeaway: GPT is amazing at the popular, forgiving languages. But when you push it into niche, brittle territory, it stops being a rockstar and starts being that intern who promises they “totally know Assembly” but secretly has Stack Overflow open in the next tab.