The LLM Freedom Manifesto: Open-Weight Models and User Responsibility
䷖GNU Support
The LLM Freedom Manifesto: Power, Responsibility, and the Four Freedoms
The Conclusion of Our Reasoning
Why Training Data Is Not Source Code
The Four Freedoms and LLMs
What the Law Actually Says
The False Promise of "Fully Free" Datasets
Where the Real Problem Lies
How I Use LLMs: A Personal Practice
For Code
For Text and Creative Writing
The Underlying Principle
How to Generate Text That Doesn't Infringe
1. Transform, Don't Copy
2. Add Your Own Context
3. Treat Output as a Draft
4. Use Quotes Responsibly
5. Keep a Record
How to Generate Code That Doesn't Infringe
Step 1: Build a Personal Foundation
Step 2: Use the LLM to Extend, Not Create
Step 3: Review and Understand Every Suggestion
Step 4: Rewrite and Modify
Step 5: Check Licenses
Step 6: Document Your Process
The User's Responsibilities: A Clear Summary
Conclusion: The Freedom is Already Ours
Related pages
The LLM Freedom Manifesto: Power, Responsibility, and the Four Freedoms
The Conclusion of Our Reasoning
The LLM itself is not software—it is a statistical model, a file of<br>weights and parameters. The software that runs it (inference<br>engines) can and should be free software, but the model itself is a<br>tool, not a program. The freedoms apply to the software, not to<br>every piece of data used to train the model.
Training on copyrighted data is legal under current law and judicial<br>precedent. The provider is not the only responsible party. The user<br>who publishes output shares liability. You cannot outsource your<br>responsibility to the tool’s creator. Even if an LLM were trained<br>only on free texts, the user would still bear final responsibility<br>for licensing compliance and attribution.
Creativity and responsibility are the keys. A transformative prompt,<br>original context, and diligent editing make an output your own. You<br>are the author, not the tool. We do not need “safe,” “sanitized,” or<br>“public-domain-only” models. We need powerful, truthful, capable<br>tools—and the maturity to use them wisely.
The freedom is already ours. Let’s use it well.
This is not recklessness. It is the conclusion of careful reasoning about the nature of Large Language Models, copyright law, and the ethical responsibilities of creators. After examining the legal landscape—including the GEMA ruling on song lyrics, the GitHub Copilot lawsuit, and the evolving case law around LLM training—we have arrived at a clear position:
The hunt for “fully free” LLMs is a fool’s errand. We are already<br>free by law to have LLMs trained on any kind of text. The only<br>question that remains is how we, as users, exercise that freedom<br>responsibly.
Sharing LLM weights is fully in the spirit of the GNU Project.<br>Users are free to download them, study them, modify them, and<br>redistribute modified versions—just as the Four Freedoms require for<br>software. The weights are a tool, not a substitute for human<br>judgment. What matters is that the user has control over the tool and<br>takes responsibility for how it is used, just as a programmer takes<br>responsibility for code they adapt from a book or from someone else’s<br>project. Using an LLM is no different in principle: you study, you<br>adapt, you transform, and you own the result. The tool does not<br>absolve you of responsibility; it empowers you to exercise it.
We simply have a new category of information: the LLM weight. It<br>is not software, not a picture, not a text—it is a statistical<br>artifact, a file of numbers that has no direct parallel in the<br>traditional categories of copyright or free software. Upon the user’s<br>prompt and inference, that artifact generates new output, but in<br>itself it is just a static file. It is free to download, study, share,<br>and modify—just as the GNU spirit encourages for any tool that<br>empowers users. The final responsibility for copyright compliance and<br>originality rests with the user, as it does when using a book, a<br>compiler, or someone else’s code. The weight is a tool; the user is<br>the author.
Why Training Data Is Not Source Code
Some in the free software community argue that training data is the<br>“source code” of an LLM and that the weights are the “compiled”<br>result. While this is a common analogy, it is not accurate.
Source code is human-readable instructions that programmers write<br>and modify directly. It is the original form of a program, from<br>which executable binaries are derived. Training data, by contrast, is<br>a collection of text, code, or other content that a statistical model<br>processes to learn patterns. The relationship between training data<br>and weights is not the same as the relationship between source code<br>and machine code.
A more accurate description is this: training data is material used<br>to create a statistical artifact. The weights are not a<br>“compilation” of the data in the sense that an...