Other ones being to keep paying the AI bros in perpetuity, and having all your prompts, data, and IP getting stolen by the same people with the goal of eventually undermining your own ability to think. We will all be as free as the people living in Pods in Matrix.

"Being dependent on a small number of AI vendors, who also penetrate every aspect of your life to control, report, and spy upon you, forced to use AI at work or be laid off, or just be made redundant and let off anyway, while drowing in slop from every corner of social media, news, youtube, search results, and any other cultural production -- that's real freedom!"

I'm also not sure how that ended up the definition of real freedom. Maybe some Spinoza is in order but genuine freedom is proper understanding of one's own, and others, nature and causes. That is to say one is free to the extent that one has adequate knowledge.

To live in ignorance, that is to be pushed, pulled or compelled by what you don't understand, for example because you've outsourced your life to the magical mystery programming box, is in his terminology bondage.

Which also explains why all these AI psychosis people sound like crack addicts. Stallman is still completely correct about what software freedom is

I would guess that for most people AI still promotes this definition of freedom.

The modern world is too complex for anyone to understand all topics fully. So as a result of this cognitive cap, by Spinoza's definition people have grown progressively less free for centuries. AI can, for anyone who seeks it, allow easy access to fields where they previously would have to consult with a human expert in order to begin to understand. Coding agents arguably make coding more accessible then it's ever been, it's just that as a side effect they make understanding less necessary to produce software (and thus also less valuable). Any schmuck can now ask an agent "hey how does this program work" and get a competent answer in terms they can parse. Of course, most people do not do this. Human beings, as a general rule, do not do things they don't have to. But the option is there.

As a board saturated with experts, people here on HN don't tend to appreciate what kind of incredible super power vibe coding is for people who did not pursue an education and career in software engineering. Everything that a computer can do is now accessible to everyone. As the vast majority of people on earth would not and could not spend the time and effort to understand software (or simply don't have the talent to do so), this is a pure upgrade in terms of average freedom.

You don't understand Stallman then. Now you are dependent on closed source from closed source AI vendors. True software freedom would then being able to dissassemble the bot writing your code.

Even if the model was open source, nobody would still know how the generated code works. You wouldn't even really know in want interesting sense of the word why the model produced the output you got.

LLMs are even more proprietary than regular proprietary software, I think you've got it right here. They are destroying the free software movement and it seems like most people don't even realize what's happening.

LLMs are a fundamentally different category of thing to software. Developing them is closer to breeding cows than writing code.

Software and slop.

> You don't program an LLM

For now. But it may well become common practice to construct weights directly, or at least to directly adjust weights in trained models, rather than exlcusively train them from prior datasets.

The damage isn't restricted to software. AI is the intellectual equivalent of asbestos.

Asbestos is an amazing fire retardant. That's why nearly every building built before we knew how cancerous it was contains it.

Open source really should mean open training dataset, not merely open weights. There are some truly open source models and you can reproduce their training process if you have the compute. I can't think of a reason why those kinds of models would be incompatible with Stallmanism. Presumably they will get better and more widespread over time, like free software

Here's an example of a model that I believe fits this description: https://huggingface.co/common-pile/comma-v0.1-2t

Actually, this one also excludes nonfree data from the training set; there are probably more capable models that include downloadable nonfree data. I'm not sure if the dataset also needs to be free in order for the model to be considered free.

No one knows how the open source models work either.

It's clear that without hardware freedom, even if we disassemble the bot, the means to run it can still be controlled by hardware vendors. We need our own means to make the substrate for the intelligence too. But the problem is, as long as the manufacturing process is impossibly complex, and relies on minerals extracted all around the world, requiring complex supply chains, its beyond any local group of people to do themselves. What we need is a way of creating substrate for open intelligence just form the materials and resources that are available to any human, and then the freedom to train it to behaving in a way that alignes with our own values and local situation. But the problem with OSS adoption, is that its a real chore for normal people to set up and use it, so we need to make this process of creating our own intelegence host as fun as possible, to encourage people to do it..

This. Nailed it. This right here is the core issue. Thanks for writing.

> What we need is a way of creating substrate for open intelligence just form the materials and resources that are available to any human, and then the freedom to train it to behaving in a way that alignes with our own values and local situation.

I can't tell if you're being cheeky, but... you're describing human child-rearing. From food and water and two humans, substrate can be created, and parents, schools, and other community resources help to train and align.

I agree. We are locked into GPUs and RAMs for now at the mercy of Hynix and Samsung and Nvidia.

The next best alternative is to work on making this happen on older gen and existing hardware that we have. Current VC and SV is too focused on just going wide instead of optimizing.

Nobody knows how Stallman, or anyone else, works either. But folks don’t need to understand how he works to understand, say, EMacs.

Folks can take EMacs as is and decipher it in isolation of Stallman, or any of the other authors.

But even then, how many people understand the internals of EMacs, or the interrelationships of the zillion lines of code in hundreds of files? Yet, millions of ignorant people leverage and apply EMacs to their daily tasks.

Look at what the REA impact is having on all sorts of code bases. The ONE THING that pushed Stallman over the edge and onto his path. The lack of source code to a print driver (or something along those lines).

Well, it’s all Open Source now.

Not only is the source “open”. It can be explained.

“Here is the source tree to EMacs. ELI5 the garbage collector.” “Working…”

It goes both ways. Now everyone has access to a savant to explain software, if that’s what they like.

A savant who confabulates about 10 % of the time, per the best benchmarks? How do you trust it?

How do you trust a human? Are humans infallible, or at least better than AI at explaining things? I known my answer.

I trust the Emacs contributors, the actual experts in the topic, to fact-check their own answers, because they are incentivized for doing so (reputation).

Have you ever worked with a modern agent? They let many weak assumptions slip because their cognitive functions aren't remotely as deep as human's, but they have also been trained to hyperfocus on verifying and grounding everything they do and say, you have to actively stop them from doing so when it's unnecessary. This not what what I can call confabulation in any good faith, let alone in 10% of cases. I think you've been misled by the old paper you cite below, it doesn't say what you likely think it says.

Yes, I mainly work with OpenCode, with GLM as an orchestrator agent and DeepSeek Flash as a worker agent. I’m assuming it’s modern? I don’t have the money to use either Fable or Astra, but my assumption is that the few more points they get on the benchmarks still don’t make up for the wild confabulated things they do in my projects.

I can't speak for the other providers but Gemini carbon feels like the difference between gpt3 and gpt5 when compared to Chinese open models.

Both are fine, though not frontier of course. They mostly take unwarranted shortcuts, trying to find the easiest way to reach the literal goal set by you, failing to see the broader picture. But they definitely do not confabulate, as in they don't just make up nonexistent things, they would rather walk extra mile to verify when they're not certain, which is the reason the software they make is trash but it works in the narrow sense. They can be rigid in their writing style do to mode collapse, but I haven't seen this leading to them inventing nonexistent functions or anything like that in a long while. RL makes them do all this, modern LLMs are very far from being next token predictors trained only on text.

> But they definitely do not confabulate, as in they don't just make up nonexistent things

I've definitely seen both Labs' "best" stuff confabulate to this day.

Agreed.

The harnesses / RLHF do a good job of prioritizing mechanisms to check results, which works out pretty well pragmatically for programming.

Nonetheless, I see Opus 5.5 spew out false information daily (never mind its code's unreadability).

I can't speak to Fable or Astra, because my employers, while threatening to fire us if we don't use the bullshit bots, have also refused to light the barrels of cash on fire that are required for those models.

As far as I can see, their fundamental nature has not changed, and truth is not something they have access to - just the called tool's exit code.

> How do you trust it?

Obviously the answer is to stop relying on trust entirely, and just proceed directly to verification.

A freedom that only works as long there is an AI model one is able to reach for, with stolen data from the whole known universe.

I bet that person thinks driving a car is real freedom too.

(You need to pass a test, maintain a license from the government, in many places have proof of passing a maintenance test, pay an annual fee to the government, roads can be closed at will by the government. They can install tolls if they wish etc. In the UK, the police can stop you and search your vehicle without reason and I'm sure that's true in many other countries)