Also China has an insanely diversified grid, that utilises almost a global scale of fossil fuels.

So its like green communism at best

As an exercise so I could learn the technology, I trained a model exclusively on the Public Domain works of L Frank Baum, specifically the "Wonderful Wizard of Oz" series (did you know there are 14 books in the series just by Baum?!). The model created is not useful as a tool, and more often than not just produces English gibberish, but every now and then it can produce original coherent statements.

Again, I didn't do this to produce a useful tool, but rather an exercise to learn how to build models.

They are all trained on copyrighted material without permission, no LLM is ethical

However, this means I can refute your statement because 100% of the input data is public domain novels. Also, I trained it on my own hardware powered 100% by solar power.

The absolutely most generous copyright protection length is in Mexico which is 100 years plus the life of the author when created. So lets generously say a total of 160 years. Most of the rest of the world is 120 year max. Anything outside of that is in the Public Domain. Many things had a much shorter path to the Public Domain falling into it in as little as 20 years.

So lots and lots of stuff is not copyrighted.

Question, what do you do with it and how is the data that underlies the model harvested?

I have a friend who runs his AI locally because he doesn't want to pay for tokens.

What are the most successful/useful distillations you can run locally?

distilled from Western frontier models.

What does this mean please (remember the title of the post!) and how do you know?

It does if you want to go that route. https://www.kaggle.com/datasets is a good starting point.

Just as I suspected...

Thanks for taking the time.

there are very few that can be considered ethical - Olmo, Apertus, Starcoder

This is software meant to be run always and completely locally?

Sorry to whine, but so far nobody has eli5'd what "open-weighted" means, or "model" at that... please?

Thanks for the info!

OK but what does open-weight model mean?

Basically, it's really really expensive (from a compute standpoint, and more compute = money and energy) to train a decent LLM just using books and conversations etc. Anthropic, OpenAI, etc spend a truly insane amount of money doing this is in data centers.

Running an LLM like that is also extremely expensive from a compute standpoint, and right now the true frontier models can basically only be run in datacenters.

However, once you have a really good LLM, you can use it to train another LLM. Because it's basically already refined all its original training data (and is capable of further refining it'd output on the fly), training a new LLM from it is vastly easier and cheaper.

Additionally, you can use a much smaller and more efficient model, so it can run on less powerful hardware.

So all local LLMs that are any good that exist right now, were trained using the outputs of existing super expensive frontier models. They could not exist without the frontier models to train them. This is also how OpenAI and Anthropic create their cheaper models. Mythos/Fable is anthropics current frontier model and they distill it into (use it to train) their cheaper models like Sonnet and Opus.

Thanks. What are quantized models?

Is it really feasible to run it 100% locally? I know there’s plenty of people with very powerful rigs indeed, but still. Or are 99% of these people really saying “it would, in theory, be possible to run that locally, therefore your concerns are invalid”?

Very useful 35B models need (more or less) 8GB of VRAM and 32GB+ CPU RAM to be usable locally. 16GB RAM might work on a lean system with an RTX Nvidia card and an exl3 quantized model.

I know we are in a RAM apocalypse, but pre-apocalypse, that’s a quite reasonable requirement, IMO.

Personally, I run Deepseek V4 07-31 Flash at 19 tokens/second on a desktop with a single RTX 3090 and 128GB CPU RAM, and that’s an extremely capable model. Again, that’s expensive these days, but pre-ram apocalypse, that is not an unreasonable workstation/homelab.

If yes to the previous: the software doesn’t come from nowhere and ultimately still relies on gas-turbine-powered datacenters…

Most open weights LLMs are Chinese. And they are:

• Trained on pennies. They have to be, as they simply do not have a sea of GPUs like Big Tech. Training costs for their large models are in the millions or tens of millions; a single steel forge has used more energy than all of those training runs combined.

• China relies more on renewables, and I believe the datacenters aren’t so hastily constructed with gas turbines in the middle of cities.

• And as of now, they are transitioning away from Nvidia GPUs. Some labs already use Huawei accelerators.

…and stolen IP and stolen personal data?

Yep.

This is a huge caveat.

You can avoid this. Nvidia Nemotron models, for example, are trained on completely open datasets you can download and inspect yourself: https://huggingface.co/nvidia

They are very good, but just behind state of the art.

But in practice, the SOTA models most run use private datasets. Lord knows where the Chinese get it from, but given some common quirks between models, at least some data sources are shared (and possibly government provided?)

…However.

I would argue providing the result of the training as Apache licensed weights counts as “fair use,” in the same way non commercial fan works do.

They aren’t making a dime off releasing those weights. I’m not trying to sell anyone anything when I use them. Where is the IP theft if money isn’t changing hands?

Now, the Chinese LLM services they charge for? I have no excuse for that. Once money is on the table, it is definitely IP theft.

Question, what do you do with it

I use it for both fun throwaway stuff as well as productivity tasks for learning.

Example of fun throwaway:

Do you remember that episode of Seinfeld where the Kramer starts using Facebook marketplace and starts buying the most worthless items before being robbed when trying to get a too-good-to-be-true sale? No? Because it never happened, but you can plug that premise into an crafted LLM prompt and it will pop out a whole TV script with in-character dialog for each actor as well as use of popular existing sets.

Foreign language learning:

I'm studying a foreign language and want to interact with just the level of vocabulary and grammar I have knowledge of right now at my level for practice. I can prompt the LLM to limit itself to just what I know now and adjust the conversation level so I can practice. If I ever get stuck, I can ask the LLM to explain the grammar usage or vocabulary choice.

and how is the data that underlies the model harvested?

I'm not sure what you're asking here. Are you asking, for example, how the Deepseek model was trained? If so, I answered that above. If not, can you rephrase your question?

It means the weights (the numeric data that makes up the model) are publicly available to download and use for free.

In some cases there are conditions such as, if you run the model as your business, you need to purchase a license. That's usually the largest, most powerful models, though, not those most people would be able to run at home.

You run a program like llama.cpp that can use the weights to run the model, which you can then use for whatever you'd use a model for: figuring out tech stuff, coding, etc. It's a bit involved, but there are tools that make it easy to get started. LM Studio for instance.

In some cases, the lab that created the model publishes the dataset that it was trained on, and those are usually made of publicly available data. In most cases, though, the labs don't give details, but the answer likely involves siphoning every web page they could find.

"Weights" in LLMs are the "final answer" numbers from the results of model training, and these are the engine of the LLM model. Lets use chocolate chip cookies as an analogy.

The chocolate chip cookies are produced with ingredients, a recipe, labor effort to produce the dough, and then cooking energy/effort to bake chocolate chip cookies. In a traditional AI company, the company gets the ingredients, they write their own recipe, do all the dough creation, and then the energy/labor for baking, and charge you money to get chocolate chip cookies.

An open-weights AI company got all the ingredients, wrote their own recipe, labor effort to produce the dough, and instead of charging you for the dough, you get as much uncooked dough as you want forever. Your only task is to take the dough and cook it yourself and you've got free chocolate chip cookies. You can make as many cookies as you want with your own oven. However, you are not given free raw ingredients, nor are your given the recipe to alter it in a way you might like. You can only get the dough for free.

So open-weight AI models (uncooked dough) are LLMs you can use on your own computers (oven) for free and have LLM output ( chocolate chip cookies), but you don't get the training data (ingredients) nor the training parameters (recipe) that built the open-weight model.

careful, you might rage-quit soon

Ollama for text and multimodal (image to text). It supports structured JSON responses you can tie un to cool flows.

ComfyUI for image/audio/video/3d model generation. Use a docker image because installing and configuring it raw is a nightmare.

Huggingface is a good spot for the models. Ollama has its own library. There are several others.

You can get basic text interactions and small images on a typical office PC. You can make nice enough stuff with a typical gaming PC. More power definitely means better results though

*CivitAI also great for models, and they have a .ted domain too

Wow. as if the original models weren't rickety enough.

There is Apertus which at least claims to only use permissively licensed sources and respect robots.txt opt-outs. They outline their methodology and source datasets in this document. Though I haven't checked the actual sources myself.

Thanks, that was properly eli5'd!

Thanks for the clarification.

Where is the IP theft if money isn’t changing hands?

Money not changing hands is pretty much the definition of theft.

Kind of like "compressed". It takes longer/more effort to run them, to produce the same result as a the same model that has not been quantized, where that non-quantized version would consume significantly more RAM but produce the result faster. You would typically only run the quantized model when you're starved for RAM, which most of us are running LLMs locally.

Think like zipping a file with file compression. It takes less space, but has to be unzipped for you to have usable files again.

completely open datasets

Not completely -- it is mostly open, but they use a dozen or so private datasets for things like training on global regulations, minesweeper (for some reason), etc. To their credit, they do indicate this on the model cards, but it's not entirely clear what is in those datasets either.

(I fell for that bit of marketing myself awhile back.)

Borderline strawman there but I'll bite.

Open weight models trained unethically are unethical. Closed weight models trained unethically and then sold back to you for a profit from gas-powered datacenters funded through Ponzi schemes are substantially more unethical.

From there it's a harm reduction calculus. No, those are never pleasant.

So do you let the closed weight labs conquer the field unopposed just so you can feel better about yourself? That's a valid stance, FWIW, and it's also super easy and convenient because you don't have to do anything. It especially makes sense if you believe it's still possible that AI will just go away on its own. I don't, myself, not anymore, so I encourage the use of open weight models, however grudging, so people don't give money to the closed labs and in the worst case aren't eventually stuck with the maximally unethical options. And we've not even touched on the nightmare labor replacement scenarios that seem every day less unlikely. Am I right? I have no clue. Like you, I'm just trying to make the best choices I can in a world that's gone to shit. I'd recommend dropping holier-than-thou attitude either way, though, because it doesn't help our side. Man.

I'm going to simplify a little here, so don't take this completely at face value.

Models are, quite literally, long series of numbers (called weights). A model might store the weights in 16 bit numbers (that is, 16 binary digits). The size of the model (and how much memory it needs) is determined by how many weights there are, and how many bits each weight takes. You can take a 16 bit model and rework it to use 8 bit, or even 4 bit numbers. The result intuitively behaves a lot like the same model, but with less precision to the weights. That makes the model take way less space in ram, but also makes it more likely for concepts (encoded in the weights) to overlap, which impacts model quality. Often the effect is that fine distinctions get lost.

They're honestly not that rickety. If you're getting your news from social media (including Lemmy) you may want to actually try them rather than just take the random incidents where things go wrong as broadly representative.

As for the power/environmental impact, for what they do LLMs are actually very low impact per-request.

Worth noting that a request is often dozens of requests now that there's "reasoning," even for search. As I understand it, a model will take a question , figure out the context (one request), reform the question so it yields better results (another request), if it's doing a web search there's requests for each result, another to compare, another to check if it answers the original request, if not it loops and does it all over again. So one request is easily, and often, dozens of requests. This is one way Ai companies can say to investors, "see, look at how much usage has increased."

Also, we need to factor in the power to scrape and train each of those models, build the datacenters which is near impossible as these companies are not transparent about it and actively try to obstruct investigations into it. Then there's the redundancy of all these different companies competing and doing roughly the same thing at the same time, as fast as they can, so it's orders of magnitude inefficient energy consuming before it gets its first user prompt.

Comparing it to other assaults on the environment is not only hard to do, but a case of "the worse negates the bad" fallacy.

Unlike the definition of heft which is a photo of OP’s mom.

I do see what you're saying. You can account for all of these factors and it still turns out that, largely, individual LLM use just doesn't use that much power compared to most things people do day to day. Inference is so cheap that even dozens of requests don't amount to much. I could look up and give you a bunch of numbers, but I don't think that's likely to convince anyone who doesn't do the research themselves. It's so easy to come up with sources that say what you want. I'd encourage you to actually look into this yourself.

Training costs are higher, but you train once and use repeatedly. Right now, total training costs are stupidly high, but that's because we've got an arms race between the frontier labs to spend as much money and compute as they can for truly marginal gains in quality. The solution to that problem isn't for individuals to stop using AI, it's to stop those assholes from wasting so much power.

Individual LLM use is so cheap, that it really isn't worth wasting people's energies thinking about limiting that. Instead of being distracted by attempts to make this an issue of personal responsibility, we should be focusing on what will actually make a difference. We should be focused on supporting policies that lead to systemic change. A carbon tax would change corporate behavior right quick, and not just for AI companies.

What I'm saying is it's akin to writing a fanfic or making fanart of your favorite franchise. Or getting inspired by a painting you see, and making something similar yourself.

Do that for your personal enjoyment? That's fair use, under the law.

But the moment you start trying to sell it is when you get in legal hot water, and when it's indeed morally problematic.

The scale is different, but I'd argue a similar principle applies: if you use some model trained on public works from a protected IP, but the model and its outputs are not resold, nor profited from, it's not theft. The point is beyond money not changing hands; there's no profit being made from the original author's stuff. They aren't being taken advantage of any more than someone viewing their public stuff for free, or someone creating derivatives from private work.

But all that is off the table the moment profit and distribution is involved.

Cool that you did that. However, "Large" as in "Large Language Model" generally refers to the dataset size, and 14 books is not Large.

There's a lot of info that you need to know to explore this space, so I'll take my own shot at answering. Let me know if anything needs further explaining!

An "open weight" model is an AI model where you can download the data needed to run the model on your own hardware for free. Contrast this with proprietary models-as-a-service like ChatGPT and Claude where you have no access to the data needed to run the model yourself -- you can only use it through the services provided, usually for a fee, and which can be taken away from you or changed at any time with no recourse.

The mapping to traditional open source terms does not work well since what you get is a binary artifact.

Those artifacts are released with a license -- and many of the models are licensed permissively (e.g. MIT or Apache license terms). You can take those weights, modify them, and then release them as new models -- and people do actually do this in practice!

Is it really feasible to run it 100% locally?

Yes. I run models on my own computers and have tried a number of configurations to figure out what works well. The Qwen family of open weight models (from Alibaba) are the ones I've found most useful so far. Gemma4 models (from Google) are also useful.

I prefer models that have been modified by the community to remove corporate censorship -- i.e. stripping that "As a large language model..." cover-your-ass crap and evasiveness on topics like Tiananmen Square. If that means the model is technically capable of telling me to go kill myself too, so be it; I've spent 25+ years dealing with assholes on the internet and can handle abuse from a stupid robot if I have to. (In practice though, they're usually pretty nice still unless I deliberately tell them to act like an asshole -- and then Qwen, at least, starts to sound like a snarky redditor; it's quite funny most of the time, actually.)

If yes to the previous: the software doesn't come from nowhere and ultimately still relies on gas-turbine-powered datacenters and stolen IP and stolen personal data, no?

Models do require training to create, yes. It's generally not clear where the training physically happened IRL -- so, yes, some of them probably used power from gas turbines, but others may be drawing power from the Three Gorges Dam in China or solar plants or nuclear plants or whatever else is hooked up to the electric grid where the training happened. Most of them are also not very open about the data sets they were trained on. (There are exceptions to this though!) The Chinese models in particular are almost certainly trained heavily on logs extracted from Western models in addition to using whatever other data they could get ahold of. Whether you think that's ethical or not is a matter of perspective; how do you feel about Robin Hood?

Once a model has been trained though, it can run on a normal GPU. The power requirements to run an LLM are basically the same as running a video game, or, equivalently, about the same as turning on a few incandescent lightbulbs. (The iGPU in one of my systems uses 100W; the discrete GPU in another system I've tried uses 215W under load with appropriate tuning -- or 300W if you run it naively.)

If you want to run a model yourself, I recommend using llama.cpp -- there are instructions on how to get started with it here: https://llama.app/

These are the models I've found most useful:

• Qwen3.8-27B (official version): https://huggingface.co/Qwen/Qwen3.8-27B
• Qwen3.8-27B (uncensored): https://huggingface.co/MuXodious/Qwen3.8-27B-absolute-heresy
• Qwen3.6-35B-A3B (official version): https://huggingface.co/Qwen/Qwen3.6-35B-A3B
• Qwen3.6-35B-A3B (uncensored -- ⚠️ mildly NSFW graphics on page): https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-heretic
• Gemma4 (official collection -- many variants in different sizes): https://huggingface.co/collections/google/gemma-4

If you have an iGPU only, I recommend using one of the so called "Mixture of Experts" (MoE) releases. These are typically named like 35B-A3B or similar; the first number indicates the total number of weights (35 billion) and the second indicates how many are "active" (i.e. actually used during computation) at one time while the model is running (3 billion in the example). These models need less computation to run and stay fast on weaker GPUs. Qwen3.6-35B-A3B is very good in this space and was my go-to model for a long time.

If you have a discrete GPU and enough VRAM, I recommend using a dense model (i.e. one that activates all its weights while answering) like Qwen3.8-27B.

It's worth noting that people don't usually use the full quality weights (which are typically ~2 bytes per weight); they use a "quantized" version -- compressed in a lossy fashion like a JPEG. Going down to 4-bits (half a byte) on average per weight is about as low as most people like to go -- you will see this indicated in names like Q4_K_M. (Quantized to ~4 bits with the K quantizaation scheme, medium variant.) Usually a bigger number is better in the sense of "closer to the original quality" -- at the cost of needing more RAM.

Full quality weights are often found as safetensor files on HuggingFace. Quantized weights intended for use with llama.cpp are usually in GGUF file format.

Does that help?

My workplace. Others by the same company.

But it is AI.

true, but that's hardly relevant as a reply to s comment which is clearly referring to LLMs and not machine learning in general

I’ve heard good things about Qwen.

This is software meant to be run always and completely locally?

That's what is claimed, I haven't run them locally since I don't have a good system.

To be honest, I'm not sure if I can eli5 weights and models, but I'll try. Think of a model like the base - for example, OpenAI has different models like Astra, Sol, etc. These are different models, like different versions of a software or operating system like macOS, but for AI stuff. Like one would download a software, you download a model to perform tasks.

Weights are vales that can influence inputs of these models to get a desired/better result. What most of these models are doing is mostly predicting what might be the next appropriate text/data to the question you asked. When you ask these AI models what 2+2 is, it is not performing a math operation like a normal program, it is looking at its training data to see what the closest option might be. It is doing pattern matching.

These AI models inside can be thought of like an interconnected network, like neurons in our body, that keep passing information to the next neuron and to the brain to make a decision. (Before understanding LLMs it would help to understand Neural Networks first). These AI networks need weights and biases. These networks perform calculations and weights are used to determine how much importance/weight each input can have on the output. Bias on the other hand, is used to shift/change the output so the AI model can 'learn' to pattern match better.

What open-weight models, do is they make the model available for download along with the weights. No information is given on training data. Like with ads, ones with most data emerges victorious i.e, has a better model. So these companies do theft, don't list their training data afraid of getting caught. I forgot which one, but either Deepseek or Qwen (both open-weight) was caught 'stealing' from Claude (not open weight). You can probably guess how much these companies value ethics.

I'm not sure if this entire thing goes away, but local models might be the ones left standing when this bubble pops.

Open weight is different to open source. Open Source AI as it stands, the definition requires a model to have entire thing made public - so the weights, biases, training data used, the model. Apertus, Olmo etc are mostly meeting open source AI definition.

If you need to know more, this is what we'd use to refresh our memory before exams :)

I probably might have made mistakes here, English isn't my first language either. But I hope you get an idea about these terms

If you really need to understand this tech more, I recommend watching 'AI for Everyone' course on Coursera from Andrew Ng. It is free to audit, my friends who took his course were hyped (I wasn't really interested in AI)

Yeah, real solid arguments; So rage inducing.

i agree.

in that case, wouldn't it be similar ethically if people use these models on their own machine for personal use?

Right... I meant like which specific file.

If you can play modern games, you can probably run a local LLM. The more VRAM the better it'll go.

For ethics, there's two issues. The first is power consumption. Running a model locally isn't consuming any more power than running a game that fully pushes your system. It's not really a concern.

The second issue is content generation. It's all created through the theft of content others have made. There's no getting around that. As far as I'm aware there are no ethical models that specifically have not stolen content. If you're just using it to chat with or whatever, I think it's fine. You probably aren't taking work from anyone, at least to any degree more significant than piracy. If you're generating stuff to be distributed, then you're distributing stolen work. If that's ethical or not is up to you to decide.

The comment right here I'm 99% was sarcasm. I believe they're poking fun at the original comment you replied to being incredibly stupid.

Is it ethical for me to steal someone's car? It was sitting on the street available for anyone to access. Why would it be wrong?

It's the backend stuff I wanna figure out so it can convert plain language requests and notes into personal assistant tasks while being flavorfully bitchy about all of it. Apparently RAG and agent tools have something to do with it.

Thanks a lot.

I did not know it was possible to influence AI software/models and nudge them in a certain direction.

These AI companies have even more power than I thought for a long time.

For ethics, there’s two issues.

No, there's seven main issues.