brucethemoose @ brucethemoose @lemmy.world

Posts

22

Comments

2,093

Joined

1 yr. ago

That's fascinating. I vaguely knew of the superstition angle, but not specifics or the extent.

There goes my afternoon, thanks.

But it does remind me of similar issues in other countries. China, for example (not to single them out) has issues with Eastern Medicine culture conflicting with scientific practices, right?

Yes! Fission power is objectively great, with the biggest caveats being the huge upfront investment, slow construction, and (depending on the specific technology) proliferation concerns.

I honestly though Trump would consider it 'woke' as opposed to 'clean coal,' a term he used.

Yes, but its clearly a building block of Meta's LLM training effort, and part of a pattern.

One implication I didn't mention, and don't have hard proof I can point to, is garbage in garbage out. Meta let AI slop and human garbage proliferate on Facebook, squandering basically the biggest advantage (besides cash) they have. It's often speculated that, as it turns out, Twitter and Facebook training data is kinda crap.

...And they're at it again. Zuckerberg pours cash into corporate trash and get slop back. It's an internal disaster, like their own divisions.

On the other side, it's often thought that Chinese models are so good for their size/compute because they're ahem getting data from the Chinese government, and don't need to worry about legal issues.

...Well, the pro nuclear angle is a tiny silver lining?

The research community already knows this.

Llama 4 (Meta's flagship 'AI' project) was as bad release. That's fine. This is interative research; not every experiment works out.

...But it was also a messy and dishonest one.

The release was pushed early and full of bugs. They lied about its performance, especially at long context, going so far as to game Chat Arena with a finetune. Zuckerberg hyped the snot out of it, to the point I saw ads for it on Axios.

Instead of Meta saying they'll do better, they said they're reorganizing their divisions to focus on 'applications' instead of fundamental research, aka exactly the wrong thing. They've hermmoraged good researchers and kept AI bros, far as I can tell from the outside.

Every top LLM trainer has controversies. Just recently Qwen (Alibaba) closed off their top base models just to spite Deepseek, so they can't distill them. Deepseek is almost certainly training on Google Gemini traces. Google hoards their best research for API models and has chased being sycophantic like ChatGPT. X's Grok is a joke, and muddied by Musk's constant lies about, for instance, open sourcing it. Some great outfits like 01ai (the Yi series) faded into the night.

...But I haven't seen self-destruction quite like Meta's. Especially considering the 'f you' money and GPU farm they have. They're still pushing interesting research now, but the trajectory is awful.

ChatGPT (last time I tried it) is extremely sycophantic though. Its high default sampling also leads to totally unexpected/random turns.

Google Gemini is now too.

And they log and use your dark thoughts.

I find that less sycophantic LLMs are way more helpful. Hence I bounce between Nemotron 49B and a few 24B-32B finetunes (or task vectors for Gemma) and find them way more helpful.

…I guess what I’m saying is people should turn towards more specialized and “openly thinking” free tools, not something generic, corporate, and purposely overpleasing like ChatGPT or most default instruct tunes.

So, literally exactly what was promised. In excruciating detail.

It’s mind boggling how Trumps policy is twisted positively so relentlessly. There’s so much deciphering of “oh he really means this writes an essay.” No, his platform means what it says.

Then people are shocked when it happens!

As much as history was distorted, the Nazis regime still fancied itself as secular and intellectual, right?

This one seems to view the scientific establishment as a distrusted obstacle, corrupt. There’s not even the pretense. Demolishing “woke” science is the stated point.

TBH this is a huge factor.

I don’t use ChatGPT much less use it like it’s a person, but I'm socially isolated at the moment. So I bounce dark internal thoughts off of locally run LLMs.

It’s kinda like looking into a mirror. As long as I know I'm talking to a tool, it’s helpful, sometimes insightful. It’s private. And I sure as shit can’t afford to pay a therapist out of the gazoo for that.

It was one of my previous problems with therapy: payment depending on someone else, at preset times (not when I need it). Many sessions feels like they end when I’m barely scratching the surface. Yes therapy is great in general and for deeper feedback/guidance, but still.

To be clear, I don’t think this is a good solution in general. Tinkering with LLMs is part of my living, I understand the jist of how they work, I tend to use raw completion syntax or even base pretrains.

But most people anthropomorphize them because that’s how chat apps are presented. That’s problematic.

You can still use the IGP, which might be faster in some cases.

Oh actually that's a great card for LLM serving!

Use the llama.cpp server from source, it has better support for Pascal cards than anything else:

https://github.com/ggml-org/llama.cpp/blob/master/docs/multimodal.md

Gemma 3 is a hair too big (like 17-18GB), so I'd start with InternVL 14B Q5K XL: https://huggingface.co/unsloth/InternVL3-14B-Instruct-GGUF

Or Mixtral 24B IQ4_XS for more 'text' intelligence than vision: https://huggingface.co/unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF

I'm a bit 'behind' on the vision model scene, so I can look around more if they don't feel sufficient, or walk you through setting up the llama.cpp server. Basically it provides an endpoint which you can hit with the same API as ChatGPT.

Is this an ADHD meme?

I'm afraid it might be, cause I have a trail of 'one giant playlists' and songs on repeat.

1650

You mean GPU? Yeah, it's good, I was strictly talking about purchasing a laptop for LLM usage, as most are less than ideal for the money. Laptop vram pools are relatively small and SO-DIMMS are usually very slow.

Things will get much better once the "Max" AMD SKUs proliferate.

Yeah, just paying for LLM APIs is dirt cheap, and they (supposedly) don't scrape data. Again I'd recommend Openrouter and Cerebras! And you get your pick of models to try from them.

Even a framework 16 is not good for LLMs TBH. The Framework desktop is (as it uses a special AMD chip), but it's very expensive. Honestly the whole hardware market is so screwed up, hence most 'local LLM enthusiasts' buy a used RTX 3090 and stick them in desktops or servers, as no one wants to produce something affordable apparently :/

I was a bit mistaken, these are the models you should consider:

https://huggingface.co/mlx-community/Qwen3-4B-4bit-DWQ

https://huggingface.co/AnteriorAI/gemma-3-4b-it-qat-q4_0-gguf

https://huggingface.co/unsloth/Jan-nano-GGUF (specifically the UD-Q4 or UD-Q5 file)

they are state-of-the-art at this size, as far as I know.

8GB?

You might be able to run Qwen3 4B: https://huggingface.co/mlx-community/Qwen3-4B-4bit-DWQ/tree/main

But honestly you don't have enough RAM to spare, and even a small model might bog things down. I'd run Open Web UI or LM Studio with a free LLM API, like Gemini Flash, or pay a few bucks for something off openrouter. Or maybe Cerebras API.

...Unfortunely, LLMs are very RAM intensive, and >4GB (more realistically like 2GB) is not going to be a good experience :(

Actually, to go ahead and answer, the "fastest" path would be LM Studio (which supports MLX quants natively and is not time intensive to install), and a DWQ quantization (which is a newer, higher quality variant of MLX models).

Hopefully one of these models, depending on how much RAM you have:

https://huggingface.co/mlx-community/Qwen3-14B-4bit-DWQ-053125

https://huggingface.co/mlx-community/Magistral-Small-2506-4bit-DWQ

https://huggingface.co/mlx-community/Qwen3-30B-A3B-4bit-DWQ-0508

https://huggingface.co/mlx-community/GLM-4-32B-0414-4bit-DWQ

With a bit more time invested, you could try to set up Open Web UI as an alterantive interface (which has its own built in web search like Gemini): https://openwebui.com/

And then use LM Studio (or some other MLX backend, or even free online API models) as the 'engine'

Alternatively, especially if you have a small RAM pool, Gemma 12B QAT Q4_0 is quite good, and you can run it with LM Studio or anything else that supports a GGUF. Not sure about 12B-ish thinking models off the top of my head, I'd have to look around.

Honestly perplexity, the online service, is pretty good.

As for local running, one question first: how much RAM does your Mac have? This is basically the factor for what model you can and should run.

I don’t understand.

Ollama is not actually docker, right? It’s running the same llama.cpp engine, it’s just embedded inside the wrapper app, not containerized. It has a docker preset you can use, yeah.

And basically every LLM project ships a docker container. I know for a fact llama.cpp, TabbyAPI, Aphrodite, Lemonade, vllm and sglang do. It’s basically standard. There’s all sorts of wrappers around them too.

You are 100% right about security though, in fact there’s a huge concern with compromised Python packages. This one almost got me: https://pytorch.org/blog/compromised-nightly-dependency/

This is actually a huge advantage for llama.cpp, as it’s free of python and external dependencies by design. This is very unlike ComfyUI which pulls in a gazillian external repos. Theoretically the main llama.cpp git could be compromised, but it’s a single, very well monitored point of failure there, and literally every “outside” architecture and feature is implemented from scratch, making it harder to sneak stuff in.

OK.

Then LM Studio. With Qwen3 30B IQ4_XS, low temperature MinP sampling.

That’s what I’m trying to say though, there is no one click solution, that’s kind of a lie. LLMs work a bajillion times better with just a little personal configuration. They are not magic boxes, they are specialized tools.

Random example: on a Mac? Grab an MLX distillation, it’ll be way faster and better.

Nvidia gaming PC? TabbyAPI with an exl3. Small GPU laptop? ik_llama.cpp APU? Lemonade. Raspberry Pi? That’s important to know!

What do you ask it to do? Set timers? Look at pictures? Cooking recipes? Search the web? Look at documents? Do you need stuff faster or accurate?

This is one reason why ollama is so suboptimal, with the other being just bad defaults (Q4_0 quants, 2048 context, no imatrix or anything outside GGUF, bad sampling last I checked, chat template errors, bugs with certain models, I can go on). A lot of people just try “ollama run” I guess, then assume local LLMs are bad when it doesn’t work right.