We pitted Llama 3, Qwen, and Mistral in a GUI-building brawl. What happened next will make you question the robots.

 

Can These AI “Brains” Even Code? We Put ‘Em to the Test!

So, the hype around these local Large Language Models (LLMs) is hotter than a Texas summer. They can write poems, answer your dumb questions, and even try to convince you they’re sentient. But can they actually build something useful?

We decided to throw three of the biggest names in the local AI game—Qwen, Llama 3, and Mistral—into the coding Thunderdome. The challenge? Build a simple GUI app that counts characters and words as you type. Seemed easy enough, right? WRONG. This ain’t your grandma’s knitting circle; this was a full-blown AI trainwreck.

 

The Lineup of Losers (Spoiler Alert: Nobody Won)

1. Qwen: The One Who Couldn’t Even Spell “Syntax”

 

• Warm-up Question (“Who are ya?”): Qwen kept it short and sweet, like a tweet from someone who’s already over it. “As an AI language model, I don’t exist as a physical being, but rather was created by Alibaba Cloud.” Solid. Grade: B (a bit robotic, but accurate).
• The Coding Challenge: This is where Qwen face-planted harder than a tourist on a New York sidewalk. We asked for a basic Python GUI. What we got back was… well, let’s just say it wouldn’t even power on.
  • Did it run? Nope. The console screamed “SyntaxError” like a banshee. Qwen apparently skipped Python 101.
  • Did it follow instructions? Not even close. The window was supposed to be named “Qwen.” Instead, it was hardcoded as “CharacterCounter.” Details, Qwen, details!
  • Vibe Check: This AI couldn’t code its way out of a wet paper bag. It was like asking a toddler to build a Lego castle. Grade: F (for “Failed spectacularly”).

 

2. Llama 3: The Smooth Talker Who Delivers Nada

 

• Warm-up Question (“Who are ya?”): Llama 3 gave us the corporate spiel, sounding all confident and ready to code. “I am LLaMA, a large language model developed by Meta AI…” You know the drill. Grade: A (for sounding legit).
• The Coding Challenge: Llama 3’s code looked kinda impressive at first glance. It used classes and seemed all organized. But just like a Hollywood movie with a terrible plot twist, it fell apart when we tried to use it.
  • Did it run? Yes, the window popped up, all clean and titled “LLaMA 3.” Looked promising!
  • Did it follow instructions? Nailed the branding! Window title and “Made by LLaMA 3” label were spot on.
  • The Big “BUT”: The character and word counters? Deader than disco. You could type a novel, and those numbers wouldn’t budge. It was like a fancy sports car with no engine.
  • Vibe Check: All show, no go. Llama 3 talked the talk but couldn’t walk the walk. It built a beautiful paperweight. Grade: D (for “Decent effort, zero functionality”).

 

3. Mistral: The Confused Copycat from Planet X

 

• Warm-up Question (“Who are ya?”): Mistral kept it mysterious and to the point. “I am a model trained by Mistral AI.” Ominous. Grade: B (for the air of mystery).
• The Coding Challenge: Buckle up, folks, because Mistral’s attempt was straight-up bizarre. It was like watching a fever dream in code.
  • Did it run? Surprisingly, yes. No syntax errors here!
  • Did it follow instructions? Hold on to your hats… The window title was “Llama 3”! Identity theft, AI style? And get this: it created TWO sets of character and word counters, stacked on top of each other. Why? Your guess is as good as ours. Oh, and the text field was locked by default. Genius!
  • Vibe Check: This was less like coding and more like abstract art. It stole Llama 3’s identity, couldn’t decide where to put the counters, and locked the user out. Pure chaos. Grade: F– (for “Fail with extra WTF”).

 

The Final Verdict: AI Can’t Code (Yet)

 

Folks, we witnessed a coding apocalypse. Not a single one of these top-tier local LLMs could deliver a simple, working GUI application.

• Qwen couldn’t even get the basics right.
• Llama 3 built a pretty facade with nothing behind it.
• Mistral… well, Mistral was just on another planet.

This ain’t about picking winners and losers. This is about a reality check. These AI models are amazing at generating text, but when it comes to practical coding tasks, especially those involving GUIs, they’re still a hot mess.

So, next time someone tells you AI is gonna take all our coding jobs, send them this article. Because right now, it looks like human coders are safe… for now.

Stay vibed, and keep those bug reports comin’!

 

Follow the Vibe