Manufactus

Ai/local LLM enthusiast running weird Frankenstein rigs

[Manufactus @Manufactus1]

New board, 8 channels, avx512, i7. Let's hope it works 👏

[Manufactus @Manufactus1]

All of a sudden everyone in the feed uses PS5 for local LLMs. God I love open source and weird stuff like this

[あるごす @argos_M1111]

ローカルLLM勢がPS5（の中身）の可能性に気付きつつある

[Manufactus @Manufactus1]

Is anyone focusing on digital rendering of video/photo as a product? How are the sales and what demand are you seeing?

[Manufactus @Manufactus1]

That's exactly what my local agent said after it nukes my Nvidia drivers and causes kernel panic in my omarchy os. Thanks Qwen 3.8 flash next.

[Elisha Long @ElishaDLong]

Haven't had a single deep thought all day

[Manufactus @Manufactus1]

@LottoLabs 1k decode!? It'll start answering before you even type a prompt. That's how fast it is.

[Manufactus @Manufactus1]

@Morrowmake I'll buy sand and build my own factory to make more GPUs

[Manufactus @Manufactus1]

I used to main windows, then switched to Linux a while back. What a breath of fresh air. Then omarchy dropped and literally everything changed. I fear my PC doesn't belong to me anymore, it just kind of takes care of itself to a certain degree. All 100% local. Is this freedom?

[Tom Ballard @tcballard]

My laptop is finally mine.

Every little detail is there because I wanted it. If something annoys me, I can fix it. If something’s missing, I can build it.

I open the lid and think: fuck, I love this thing.

It’s perfect for me.

That’s why I keep building for Omarchy.

[Manufactus @Manufactus1]

Gun to your head, name an old gpu that ran really hot from stock cooling

[Manufactus @Manufactus1]

Makes a quiet small farm in the middle of nowhere with peace and quiet that much more enticing. If you go for 100 agents you will probably get baseline happy at 10 agents.

[Tibo @tibo_maker]

AI was supposed to make us work less

I've never been this tired in my life

most days I have up to 10 agents running at the same time. one fixing a bug, one writing pages, one refactoring something I already forgot I asked for, 3 other I have no idea what they are doing but i suppose i ask it?

and I jump from conversation to conversation all day

compared to 5 years ago, I probably context switch 100x more

by the end of the day my brain is fried. completely toasted. nothing left, not even for the people I love

I call it the fried brain effect

thought it was just me but almost every founder I talk to lately says the exact same thing

we got superpowers, and somehow the pressure only went up. if you can ship 10x more, you feel like you have to ship 10x more. every single day

I don't have a solution

honestly I have no idea where this goes from here

anyone else feeling this?

[Manufactus @Manufactus1]

@evankinney 256gb ddr5? What is that like $350k in 2026?

[Manufactus @Manufactus1]

Could you guys tag people in this conversation who you know are avid users of the Qwen 3.8 flash next model who could shed some much needed light? I need to get to the bottom of this.

[Manufactus @Manufactus1]

From my experience, qwen models can't comprehend a flow of conversation. There is no dynamic understanding that ties each passing section in to each other. It can almost not follow a single comprehensive workflow without holding its hand all the way. This is what I mean:

1.

[Manufactus @Manufactus1]

QUESTION FOR YOU:
I am talking about maybe 10-20% of the problems I am having with qwen models. I am surprised in how many people praise this? It's trash? At least for my purposes. Running qwen 3.8 Flash Next iq3_s. How is your qwen behaving?

[Manufactus @Manufactus1]

4. Coherently following instructions based on user demand is near non-existent. Even if I create a list like this:
https://t.co/CQ75y4cWVq this
https://t.co/f0sykt0c4q that
3.edit this
etc
it will completely fail in creating an internal hierarchy of tasks to be followed

[Manufactus @Manufactus1]

3. Qwen does not understand why a video script is good or bad. Flat out, it has literally no idea why one script is superior to another. Its like it has 0% training data on this.

[Manufactus @Manufactus1]

2. I gave it a task. create X amount of skiss-art photos and use them as loose reference when creating the real images, (here's more criteria regarding photo quality and background location) the agent fails and uses skiss art as primary photos as base canvas instead of loose insp

[Manufactus @Manufactus1]

1. When talking to a person, the flow of conversation is understood by coherently and naturally connecting pieces of passing information to build a complete understanding. The qwen model forgets after a couple of prompts

[Manufactus @Manufactus1]

@CaryPalmerr Looks like a perfectly operational station to me. Just needs a nice pillow to sit on!

[Manufactus @Manufactus1]

@unbug Surprisingly capable cards for their age but so many problems, so much heat any power consumption. They look sick though.