Hacker News Reader

Comment by seba_dos1 | Hacker News Reader

Compiling Android takes many hours and requires a lot of RAM and disk space, more than I have available on my machines right now. For some things there, binary patching may actually be an easier way to change stuff than compiling the whole thing from source.

Now scale that up and consider - at which point such project would start being "FOSS" in your book without actually providing its sources on an appropriate license?

The intention behind "preferred form for modification" is to put you as a user on the same level as the copyright holder. This construct works well in a world where compiling is cheap; where it isn't, it may require some refinement to preserve the intention behind it. The copyright holder could decide to modify the learning set before pressing the "start" button, you can't.

View on HN

> Now scale that up and consider - at which point such project would start being "FOSS" in your book without actually providing its sources on an appropriate license?

At the moment when the copyright holder stops ever recompiling the code from scratch and starts just patching binaries.

We are at that point with LLMs.

> The intention behind "preferred form for modification" is to put you as a user on the same level as the copyright holder.

Exactly. And the copyright holders for these LLMs do not ever "recompile". They create brand new works that aren't derivatives at all, but when it comes to modifying the existing work they invariably fine-tune it rather than retraining it.

> At the moment when the copyright holder stops ever recompiling the code from scratch and starts just patching binaries.

So when the copyright holder considers the work done and stops changing it at all, it's now FOSS too?

I'll repeat myself, as you ignored the important part:

> The copyright holder could decide to modify the learning set before pressing the "start" button, you can't.

Even if the copyright holder does not intent to retrain their model, you are not in the same position as them. The choices they made at initial training are now ingrained in the model, putting them at an advantage over anyone else if you can't inspect and change those choices. They got to decide, you did not. Your only option to be in a similar position is to start from scratch.

Why does open source have to mean that you, the user, are on equal footing with where the copyright holder was before they even started? Where is that written?

If you wanted to write a project in Rust you would have needed to be there at the beginning, too. Same if you wanted to make it a web app versus native. There are dozens and dozens of decisions that can only be made at the beginning of a project and will require completely reworking it if you're receiving it later.

If a project needed to put all future users on equal footing with where the copyright holder was at the beginning of the project in order to be open source, there can be no open source. The creator of the project invariably made decisions that cannot be undone later without redoing all the work.

> Why does open source have to mean that you, the user, are on equal footing with where the copyright holder was before they even started?

Equal footing is the whole idea behind Free Software. Open Source puts accents somewhere else, but in practice it's essentially about the same thing.

> If you wanted to write a project in Rust you would have needed to be there at the beginning, too.

I can take the source code, inspect it and rewrite it line-by-line.

I can't take a closed model, inspect the data it was trained on and retrain it, even if I had enough money and resources to do so.

Whether the copyright holder will ever want to retrain it themselves is irrelevant. They have the information needed to do so if they wanted, I don't.

> Why does open source have to mean that you, the user, are on equal footing with where the copyright holder was before they even started? Where is that written?

It's written on the OSI page about license approval:

""" The license does not have terms that structurally put the licensor in a more favored position than any licensee. """

https://opensource.org/licenses/review-process

Reply on news.ycombinator.com