Comment by AJ007

15 days ago

In F-Droid's case this is absolutely a regulatory reaction -- this is directly related to the DMA (and to some extent, the Epic lawsuits.) Google does not want third parties bypassing Google in any way -- which probably ties in to the whole AOSP thing.

> How much longer will free and open source be tolerated?

I don't think they have a choice. Imagine what would happen to Google if half their software stack was Oracle and the EU had backdoors in to all of the management and CEO's devices and private communication. Why not use Chat Control to verify that they are complying with the spirit of EU law? Turn on the remote microphones while they are at it too.

On one hand we can lament the death of open source. Yet, open source has never been healthier. There has never been more open source software available to use and in development. Even when in it comes to AI, the best open source models are actually really damn good, better than anything that existed roughly 12 months ago. As much as Google, Apple, and Microsoft want to force you in to their closed ecosystems they fear being locked in to their competitor's closed ecosystems even more!

This could be a 10 page comment, but yes, the regulatory environment is a real threat to open source and the open internet in general. Most of those threats have been coming from the EU, with things like Chat Control and PLD. Which is unfortunate, because the future of the free world will rest entirely with the United States (Also possible that the EU will be dissolved, the monetary union will have a very difficult time during the next financial crisis.)

On the other hand, software developers and users, have become too reliant on Android which is functionally a fake open source project now. I can't think of a stronger incentive to stop Android development than telling them you can't develop here without paying us.

Calling downloadable weights and biases open source is like calling compiled binaries open source.

  • It is more like the assembly dump generated from the source code with maybe some symbol information for the functions. The download licenses are also quite limited.

    The full text training data isn't really shareable though. Since it is copyrighted when it comes to plebs like us reading them.