Hacker News new | past | comments | ask | show | jobs | submit
skohan | on: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
I've been coding using the LLM server in my living room for the past few weeks, and I haven't had this much fun with tech for ages
exe34 | parent
Can I ask, do you feel the pain of the level of abstraction? I haven't tried local in a few months, but last time I tried, I felt like I was directing a coding exercise - whereas with a frontier model, it feels more like directing a product building. "I need this feature", vs "write code to do this in this file".
skohan | root | parent | next
> write code to do this in this file

I haven't had to micromanage to this level. I usually start with a spec for a feature, which will be as detailed as I am opinionated about the feature. But it's usually on the level of a high-level context, plus some key implementation details (technology choices, key requirements, maybe an interface/API specification to 80% detail), and then the project already has high-level policies documented about e.g. how to structure files within the project.

Then I do a planning phase, task breakdown, and implementation of subtasks all within the model. I do read through it, but mostly the quality is good and I might make a couple notes. Then I do a review phase, which usually picks up a couple things. I'm moving towards less manual review of results and more automation as I learn what I can and can't trust the model with.

There's definitely a capability gap vs. larger models, but honestly I kind of prefer this workflow, as I stay more in touch with how the codebase is structured.

And it's great to be able to experiment as much as I want without worrying about how many tokens I'm burning or how close I am to a usage limit.

exe34 | root | parent
Thanks for that, it's the level I like to work too - what model/quant are you using? How much vram/context and which coder?
skohan | root | parent
I'm using Qwen3.6 27B Q4, max context with pi on 32GB VRAM (although I'm testing out Glimmer on a feature implementation literally right now). Pi is great because it has minimal context added by the agent.

Looking forward to the 3.8 27B release to compare.

bitexploder | root | parent
Pi and a similar set of tools is also likely similar to the harness these models are trained on. More complex harnesses burn reasoning tokens on these small models and in my benchmarking don't seem to be able to beat Pi ever. Usually it isn't close on some tests.
rufasterisco | root | parent
Some people like it better when they direct the solution because they walk away with a better understanding of it.

This has emotional/psychological aspects (it feels less like LLMs are replacing you), as well as practical ones (overall complexity is bounded by what the dev brain can understand/grasp).

A dev work becomes more and more about reliability, signing off safe software with a litmus test: “I will be on to handle this code failure as if I had written it”.

All the above points towards keeping tight control over some level of abstractions and delegating others.

exe34 | root | parent
I find it depends at what stage I'm at with the idea - sometimes I don't want to understand it until it works, because I've wasted enough life on things that didn't do what was promised. But once I know the idea is feasible, yes I would prefer to understand the code at some level.