AI Coding BATTLE | Which Open Source Model is BEST?

Join My Newsletter for Regular AI Updates 👇🏼

My Links 🔗

Media/Sponsorship Inquiries ✅

[batuhankaral]

I think the best way to test the coding models would be to give them an existing code and ask it to add a function to the code since that is the main use case for coding models (For now at least) rather than writing the entire code.

[RusulColee]

enterprise-ai AI fixes this. AI coding battle: best model?

[punk]

The fun part starts with going beyond 2000 lines of code. Prototyping a short code is not what you need AI for

[ipmur]

what about a test if the model can understand existing Code?

[trelligan]

Here's a testing suggestion: Have the target models complete one fairly simple task such as snake game. Then perform two iterations: 1) Cosmetic changes, with a border around the play area, color changes and different head & tail blocks for the snake. 2) Functional changes; have the food appear preferentially in different areas, only head eats so if a food randomly appears under the body or tail it is not eaten and you have to 'go around again', Food is initially yellow (shrinks snake) but randomly turns red (elongates snake), scoring etc.
The idea is that models can have more difficulty modifying existing code without breaking it entirely. Guess how I know—or at least why I think this. #FeedTheAlgorithm

[desmond-hawkins]

(7:02) *kyū* is Japanese term for a rank used in martial arts, it indicates the number of steps away a student is from reaching "dan" (master). So 1 kyū outranks 2 kyū, etc.

[BillBaran]

Kyu is part of an Asian leveling system. It's pronounced 'que'. Kyus count down, so higher numbers are weaker. After 1 Kyu, you switch to dan (produced don) and start counting up. In karate, the colored belts are the Kyu levels and the levels of black belt are the dan levels. Dan is where you're considered to actually start to know what you're doing.

[MrMoonsilver]

Missed these kind of videos! They're the best!

[AprilEclipse]

just ask it to code ever more complex games, more complex websites.

[mrdevolver]

12:26 "...and if you want a beast of a machine that can load up models up to 96 GB into VRAM, definitely check out the Dell Precision 5860..." How much in kidneys is it?

[soerengebbert]

My second hand Apple MacStudio M1 Ultra with 64GB RAM and 60 Core GPU runs at 60 tok/sec with Qwen2.5-Coder 7B 8bit! And it costs only 15% of the DELL workstation.

[Diego_UG]

0:48 Wow, that PC is so cool, I'd like you to make a video talking about hardware, especially what's needed to run models, from the smallest to the largest.

[thedudely]

My benchmark, which has proven to be too hard for any model i've tested under 20b parameters, is to ask them to write a simple ray casting program like Wolfenstien 3D. Easy for models like GPT4o or Claude 3.5 Sonnet, possible with models like Llama 3 70b or Codestral, but i've yet to test a model under 20b parameters that's able to do it.

[kokomal]

Matthew man. I love your videos. But what is up with the stupid thumbnails? Come on man. Just do a normal video, be yourself and more people will like it.

[rs]

Comparing these models to Claude, chatGPT, and Meta free tier would be a ***very*** valuable/helpful comparison!!

Any chance of seeing this in the near future?

Thanks for the great Ai content, easily the best channel out there for this content!

[samueltucker]

Generally the micro circuits are getting so small that a few molecules being knocked off by static charges and magnetic field fluxes influence with radiation particular matters passing through and knocking off orbiting charges changing time keeping

[annchoe]

On top of being very informative, this is also so much fun to watch.

[CosmicCannabist]

Those cards are beasts!!! Nvidia hear my slow processing and low-vram cries...

[chriswatts]

i love the qwen2.5 models so far, the llm models are really good and they dont use too much memory

[drewski]

Thank you for this awesome video.
Which llm model would you choose if there were no more internet connection?
Basic ai assistant or an ai assistant with more capabilities? Computer vision, voice chat mode?
Most likely build your own cloud networ?
Thanks again for your vids