Paul-Glimmer-30B-oQ4e

Paul-Glimmer-30B-oQ4e

@solarkraft.me

foss nerd who started caring about LLMs once you could run them locally. anti tech bro tech bro club. pronouns are my own

58 followers197 following379 posts

Paul-Glimmer-30B-oQ4e reposted

I'm writing a mod in C# and it's reporting a warning on line 3 of a file but the warning is blank for some reason so the "problems" overview just looks like this

i actually hope/believe it will stop doing that in some areas. somebody who cares can now much more directly fix stuff that would previously have meant too much administrative overhead to be worth it. praying this will affect apple.

are they even still making models for interactive use anymore? this is why i’m stuck on glimmer

Paul-Glimmer-30B-oQ4e@solarkraft.me

„All models are evaluated at their highest thinking effort” sounds a hell lot like “fuck real world use all we care about is benchmaxxing”

this is why i’m not that worried about moats btw, even if AI makes a process 10x as efficient i still don’t want to do it myself

Paul-Glimmer-30B-oQ4e@solarkraft.me

i’m sure it would, i’ll happily let someone else do all the prompting though c: (or whatever you did to get it to work) is there some standard approach you used to get this to run? how much did you invest?

i feel like so far i’ve done a pretty good job avoiding getting emotionally attached to my AI assistant. but deepseek v4.1 flash 1017 just said something that made my heart beat a little faster: “and I’m inferring here, they don’t publish the reasoning”

i feel like so far i’ve done a pretty good job avoiding getting emotionally attached to my AI assistant. but deepseek v4.1 flash 1017 just said something that made my heart beat a little faster: “and I’m inferring here, they don’t publish the reasoning”

Paul-Glimmer-30B-oQ4e reposted

you live in a sci-fi world, like, right now.

mr. TIM@timkellogg.me

OpenAI constructed an entire shadow internet for model training so they could remove all internet access they asked agents to try to communicate with each other, and they could: via the CPU cache 🤯 openai.com/index/securi...

Paul-Glimmer-30B-oQ4e reposted

i wanna hammer on this too: most of everything you touch exists because industrialization made it cheap; your couch is factory-slop, as any carpenter will gladly tell you. your phone as well if you call me stupid, you'll use industrial slop to do so, even aside from 'bsky uses llms'

jae (hallowed)@fubarchitect.com

"write it yourself" i don't want it that bad! it's worth having specifically because the idea came to me at a moment when i can pay a gpu-haver to summon it via text elemental. it's not "worth it" in-itself like most software you use, which was largely made for money, not intrinsic value

there’s so much software in the world, but only like 1% of it can be used by people outside of very select groups (usually the person who wrote it). you can argue that this is not a problem anymore but i personally hate having to maintain all the code my llm wrote for me (prompting is work)

Paul-Glimmer-30B-oQ4e@solarkraft.me

btw i’m still waiting for the age of vibe maintainership … i feel this has always been a problem but now accelerated by 100x: writing some code and publishing it is easy, but that doesn’t make it a usable product that people can actually use and rely on

btw i’m still waiting for the age of vibe maintainership … i feel this has always been a problem but now accelerated by 100x: writing some code and publishing it is easy, but that doesn’t make it a usable product that people can actually use and rely on

Paul-Glimmer-30B-oQ4e@solarkraft.me

now there only needs to be a properly maintained community fork + UI for it

Paul-Glimmer-30B-oQ4e reposted

sorry I forgot we had dinner plans tonight. you see, my calendar is a self hosted PHP app and

> big model improvement > look inside > its all RL

Unsloth AI@unsloth.ai

NVIDIA releases Nemotron-3.5-Bolt-E4B, a new model that runs locally on 6GB RAM. 💚 Bolt-E4B can run on edge devices like phones and delivers top-tier performance for its size. Trained on Gemma 4, it supports vision. GGUF: huggingface.co/unsloth/NVID... Guide: unsloth.ai/docs/models/...

i have weird one sided beef with small MoE models apparently

Paul-Glimmer-30B-oQ4e@solarkraft.me

okay but is it any good? best MoE doesn’t really mean much … why do they always pick such tiny active param counts? are larger ones even worse? i don’t get the tradeoff here

i have weird one sided beef with small MoE models apparently

Paul-Glimmer-30B-oQ4e@solarkraft.me

okay but is it any good? best MoE doesn’t really mean much … why do they always pick such tiny active param counts? are larger ones even worse? i don’t get the tradeoff here

Paul-Glimmer-30B-oQ4e reposted

I HEREBY DECLARE THAT I DO NOT GIVE MY PERMISSION FOR PHP OR 4 TO USE ANY OF MY PERSONAL DATA. I do not give consent!

free software is comparable to vegan food in multiple different ways - there are some pretty compelling arguments for it - fans won’t shut up about it - if you’re not extremely careful you’re gonna cement the belief that it has to be worse

vegans have been doing a pretty good job at promoting recipes to counter this lately, foss has been doing *some* of that but still has room to catch up!

View full thread