Yeah, that sound pretty usable, and seems like making the harness smarter can up the quality quite a bit. And completely agree about architecting things yourself. I find you want to basically do all the scaffolding, and lay out how the project looks like in terms of structure, the call graph, etc., and then let the model fill in the details. But also like you say, for something that's a throwaway, or a small personal project to scratch an itch, it doesn't really matter.
comments
I have nothing really to compare it to in terms of the big ones because i refuse to give anthropic/openai any money. But i have used this qwen model in opencode to port an old CUDA/Optix based project to HIP and that worked fairly well. Not an overly complex refactor to be honest because 99% is just replacing the cuda prefix everywhere with hip, but it did fix the CMakeLists.txt for me and ported the optix prime acceleration structure to hipRT as well. I also let it vibecode a react based portfolio page because i have absolutely 0 knowledge of web development and refuse to learn it because i think it is insane and that turned out perfectly fine as well.
So it is absolutely usable for real world projects and the performance is quite OK as well with ~35tok/s. If i up the context size to the full 256k window, the performance drops to 5 tok/s, which is ok, if i let it do a boring refactor over night but at that point it isn't faster than running DeepSeekV4.1-Flash-1017 through Strata and that one is much closer to the current SotA.
Definitely exciting times. We are close to being able to run very capable LLMs locally that are more than good enough for development. I am personally not a fan of 100% vibe coded projects anyway, because i like to architect things myself, so letting it go in smaller steps and/or help be with the boring stuff is absolutely fine. Unless the project is some personal stuff that just needs to solve one simple problem then vibe coding is fine imho.
Finally, I can self-host an AI model to scroll social media for me, and I can go live in the woods.
It somehow ended up on my "no javascript allowed" list in my browser. Some images are missing, but I don't get the feeling I'm missing anything important. And it's super fast!
Oh that's not bad at all. How do you find q4 performs overall?
I've been running the Ridge-GGUF version on my RX 9070 XT with ~60-80k contextsize with q4 kv cache quantization. Everything fits completely in 16GB VRAM
You'd need around 64gb vram to run 27b model at q6/8. You don't have to run multiple models at the same time though, since the same model can work all the different roles sequentially.
What is the required hardware?
Do you have the correct filter lists set? I saw some ads, bu nowhere near an inconvenient amount.
Software developers really need to start organising. I hope this kicks something off.
The sentence structure, the everpresent em-dashes, the constant rule of threes.
Could that website (404 media) be any worse?
Even with ublock it has multiple popups, most of the content is spread out past multiple different articles to try and get you to read something else.
Atrocious.
How can you tell it was written by AI?
With them open sourcing this, CLI harnesses will integrate the useful ideas in short order I imagine.
Do they have a CLI app now? Otherwise it's worthless for me.
The full text is available on (of all fuckin' things) Proquest. Fascinating paper, and has some really cool potential applications for astronomy and imaging. It's not an "invisibility cloak" for infrared cameras tho, as much as science faily may like to sensationalize that it is; the metal lattice guides infrared around the object in the center, but the lattice itself is still visible (tho it appears like it has a big hole thru the middle? Not quite clear on that)
The idea, yes. But why is this website written by AI? I don't mean the code, write all the code with AI, sure, but why is all the text AI? Someone wants to start a movement and they can't even write a basic one page about it?
How dedicated are they really if they can't say anything about it in their own words?
Replace your text editor with AI, why not, but don't replace your own thinking. That's when it becomes mindless slop instead of a tool to bring your own ideas to life faster.
Fork the 1%
I loved that
AWS is proprietary.
Isn't AWS based on OpenStack?
Yeah, I think the idea so make a 1:1 drop-in replacement which honestly would be really nice. A lot of people are vendor locked into stuff like AWS, and it will be great if these kinds of efforts start providing a migration path off of these services.
All those things already have well supported open source implementations though? Is the idea to make 1-1 drop-in replacements for people who really want that exact shape?
No hate, just dont get why would I use their auth0 clone instead of say KeyCloak.
Fuck yeah
And these types of comments expose how these people really do live in a bubble, and seem to have no understanding how most people feel about what they're doing. A certified 'let them eat cake' moment here.