@hatsunama: Local Ai upgrades in the form of new architecture #ai #news #upgrade #localai #llm

J :

Im new to this local AI stuff, I mainly use chatgpt for everything on the plus plan. How do I become informed on this side of the AI world 😖

natasharussa :

It changes nothing 😂😂

Charles Streicker :

so how do we set this up??

Maestro19 :

can it be applied to mobiles as current mobiles ate like PC's. how to set up for both: mobile & PC. thank you for keeping me on point..current...

crawford3737☯️ :

well at the same time when you do it local your ideas are your ideas and they can't be stolen you know what I mean especially when you have the unlimited free on time with it other than paying that subscription or that free time limited conversation or working something out

ToddMichaelTV :

thank you bro

ChasingGnosis :

You are singlehandedly keeping us up to date on the non-frontier world.

Mike A 🏴󠁧󠁢󠁥󠁮󠁧󠁿🇨🇦 :

Can you point me in the right direction to try it myself please?

Toad :

it seems it requires Linux and doesn't have AMD ROCm

mordwyn :

This is the most common sense approach, but may have been the first approach, but it wasn't designed to make money or be designed around a product (spark or specialized AI marketed computer) or build worldwide data centers which was the vision of situational awareness. Pulsar is closer than llama.cpp. This is significantly better than pulsar.

MG 9009 :

hmmm qwen3.8 27b runs at 3tks on my pc, if this is true it will be a game changer

Wisdom Channel :

Thank you man. What a break through! Cant wait to try it, as soon as workload allows.

300.gevaldig :

hello, which specific model are you referring to, where can I get it, thx

reinventionmod3 :

so it's Ike mac pro chips?

Midwest Misfit :

so it's like Kubernates for your AI

safeappeals :

I wanna see what this can do to improve the new hardware as well as our regular pc’s

Joe Feser :

I wonder how it runs on the mac

lghadab :

can I fit qwen 3.8 uncensored 27b on my rtx 5070 8gb vram ,16 ram ddr5 and ultra 9 275hx laptop using this freetoken

cordry :

How does it compare with Colibrí? It has a similar approach = keep experts in VRAM. This sounds a bit more sophisticated.

Rick Broider :

installing now. thank you homie. really enjoy your videos and your awesome perspective.

francoperdu⚜️🏳️‍🌈🍁⚜️ :

great, but what a totally un-searchable name

OpenclawGateway :

Wait, are you saying this is like obama split, where it split your a I across all your g p u's, but this also splits some more of it across your CPU and ram? bro, i'm about to start running some bigger models.

Thomas Jeffery :

So this is kind of what I've been working on. I've been training my experts to be specialized. I've been externalizing them so I could see when the failures happen by failures. I mean hallucinations I can see where they're coming from. I have signals so I don't just accept that hallucinations are happening. I actually am able to diagnose why they're happening and prevent them from happening in the future. can't prevent them right now but I can stop them from reaching the user which is exactly what this does by the time I'm done, it'll be about 550 billion parameters at conversational speed on a 32 GB Mac studio. I'm also getting it to work on a Mac mini 16 GB. that's the goal the goal is to have an open source model that you can trust for everybody

I_AM_SNOW :