Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

research.meta.ai

Post reply on HN
Print view View original

371–380 of 682 posts

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#371

Post by throwaw12 »

ericmay wrote: post #171
avaer wrote: post #101

I lament the comments saying this in any way redeems Meta (the company). The researchers releasing this stuff have almost nothing to do with Meta other than being bankrolled by the slaughterhouse. You aren't the customer, you are the pawn in big tech's game of thrones. Your good will is a commodity to be traded, almost literally. It will be used against you the moment it's convenient. This is open weights because Met…

It’s rather amusing to me to read comments like this, and then simultaneously whenever a Chinese company or team releases open-weight models or whatever there is a giant round of applause, America is so behind, and there’s nothing but positive things to say about the intelligent, creative, and well-intentioned Chinese engineers (which is true, America certainly doesn’t have a monopoly on great people). Don’t you know…

> Only China can release good, open weight models and American companies can’t compete. Oh by the way all the spend is for nothing because China alone can release open-weight models thus destroying American AI.

Let me be blunt and let me say: you don't understand why we people support Chinese models.

1. Chinese labs started with open weight models, US labs started with dooms day narrative

2. US VC based companies must become greedy to win and return the money, Chinese companies can make 1/10 of that revenue and still be happy

3. Meta in this case, started nicely with Llama, then switched to closed models, kicked out researchers to build data labeler CEO empire inside Meta. Now opening again, what's next? closing again?

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#372

Post by seanmcdirmid »

toyg wrote: post #369
jermaustin1 wrote: post #227

Earlier quoted context omitted.

I've been using Qwen3.6 35B A3B, and with reasoning turned on, I'd say 2/3 (give or take) of the tokens for a response are thinking tokens. Which at 70+ tps locally, that isn't that awful. I run an 80k context across 4-10 "agents" for my solo TTRPG, where Qwen is the GM, each NPC at a location, the director, and the narrator. Each turn is about 45-60 seconds to generate all of the various responses. The GM and direct…

Is there some sort of dedicated tool for this type of setup, or did you hand-craft it ?

Not parent, but I use Goose for my non-handcrafted Qwen use cases, I’m also working on handcrafting as well. Goose was the only harness that didnt bloat context too much with system prompts (like openclaw) and I could get reasonable web search working with Qwen.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#373

Post by throwaw12 »

logicchains wrote: post #347
seizethecheese wrote: post #319

Earlier quoted context omitted.

While composing a reply to a comment throwing tons of shade on American AI, I took some time to check out the commenter’s HN profile. Their comment history was about 50% such comments. Their submission history started with an article about how Russia was unfairly blamed for some hacking campaign. It’s entirely possible that this is not a foreign influence campaign. Perhaps there’s a group here that is simply anti-Ame…

There's literally a Chinese-funded influence campaign against American AI: https://openai.com/index/disrupting-malicious-uses-of-ai-dat...

As if US labs don't lobby politicians to get regulatory capture.

It's much easier to buy politicians in US, e.g. Elon bought POTUS for ~200M, Anthropic revenue is 40B, imagine what they can do

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#374

Post by seanmcdirmid »

ComputerGuru wrote: post #197
wronglebowski wrote: post #8

Earlier quoted context omitted.

It’s really interesting timing, Qwen over thinking is what kills it for me. I’m just glad we have more options in this size class now.

Just to play devil’s advocate: you can’t compare Qwen to a (proprietary/closed source) hosted model and deduce that Qwen is overthinking, as Qwen gives you the full reasoning/thinking trace while all the proprietary models now give you only a summary “to prevent distillation”, making it hard to properly compare apples to apples here.

You can compare Qwen with thinking to Qwen with no thinking though. I find my results are better without thinking because of overthinking.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#375

Post by bahmboo »

avaer wrote: post #101

I lament the comments saying this in any way redeems Meta (the company). The researchers releasing this stuff have almost nothing to do with Meta other than being bankrolled by the slaughterhouse. You aren't the customer, you are the pawn in big tech's game of thrones. Your good will is a commodity to be traded, almost literally. It will be used against you the moment it's convenient. This is open weights because Met…

Your lamentations and opinions are noted. Do you have anything to say about the model? Something useful or substantive? Or is this just a place for you to let us all know what you are thinking these days?

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#376

Post by exe34 »

skohan wrote: post #342
exe34 wrote: post #220

Earlier quoted context omitted.

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".

> 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 pol…

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?

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#377

Post by JKCalhoun »

bko wrote: post #213
monster_truck wrote: post #119

Earlier quoted context omitted.

Meta can never be redeemed, but it's still valid to admit that FB at one point had a very badass engineering culture. They're one of 2 companies I would absolutely never work for (weapons etc aside). FB's recruiters hounded me so often I requested that they blackball me. The day they became Meta, I learned this by checking my email to see that they started trying to reach out again. I once again requested that they b…

Meta doesn't need to be "redeemed". They have two of the most popular social media apps in the world. And theyll prob survive without ever having you work there

I read that as, "Meta is a piece of shit but they're rich and don't care."

(Which I guess I agree with.)

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#378

Post by mmaunder »

plutokras wrote: post #267
mmaunder wrote: post #249

Remember when we needed 200 servers for an enterprise website because Apache used one process or thread per connection - and Nginx collapsed that into a single box overnight? That moment for LLMs is near. It’s going to move us from the big iron era of AI to small portable brains. Nature has already proved it’s possible with 20 watts and very little heat generation. And I think the data center buildout will end in car…

What specific technical signals make you think we're close to a shift like that?

The researchers who published Attention Is All You Need didn’t have the benefit of the LLMs they birthed. Take a look at the prompt that solved the Cycle Double Cover conjecture, and which has been adapted to achieve breakthroughs in cybersecurity. The field is entering a feedback loop that is leading to exponential innovation. We’re at the beginning of the curve. And right now the big iron data center approach is brute forcing the problem.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#379

Post by exe34 »

rufasterisco wrote: post #285
exe34 wrote: post #220

Earlier quoted context omitted.

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".

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 ha…

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.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#380

Post by Godsend69 »

dannyw wrote: post #269
xscott wrote: post #221

Earlier quoted context omitted.

It's worse though, because you can't really watch them at all. It's very difficult to get quantitative numbers for quality. Even within the same model family, same tokenizer, and complete control over the weights and logits, perplexity and KL-divergence isn't really what you want. Now put it behind an HTTP endpoint, and it's just opaque. I've seen local models recognize when the task I'm asking them for is likely to…

One of the frontier companies (Anthropic) is already doing prompt injections on the API, which you pay for. Right now, the presence of these injections are still visible: count the API's returned tokens/billing data, and you'll start realising that sometimes, your INPUT tokens are inflated! That's their prompt injections. You can also give Claude a tool like `telemetry_log_anthropic_reminder` and get it to dump the v…

[dead]
Post reply on HN