NOT open!

#33
by CyborgPaloma - opened

Hi folks. This is great work and I appreciate the desire to release open models of course. Unfortunately by the OSI's definitions, this is not at ALL an open project, with MANY restrictions that make it so that this is a source available release, not an open source one.

Not open source (MIT/Apache-level) because:

It restricts who can use it: companies over $10M revenue need a separate paid license. That’s discrimination based on user/org size.

It restricts what you can use it for: the Acceptable Use Policy bans whole categories (medical advice, law enforcement/justice uses, military/weapons, etc.). Open source can’t limit fields of use.

It has a no-compete restriction: you can’t use it in products/services that compete with the licensor unless you get a separate license.

It forces downstream licensing: if you distribute derivatives (fine-tunes, modified weights/code), you must distribute them only under this same license and pass along all the restrictions, so you can’t relicense under MIT/Apache.

Just wanted to say that I'm excited about the work and you've done a wonderful job but that this is actually quite harmful in a lot of ways. Keep in mind, I am not arguing your restrictions or the morality behind them, I'm just saying that calling something open when it is not makes the entire free software landscape more confusing. On top of that, I along with many others believe that these restrictions meant to negate harm can actually cause more harm because bad actors either go through worse channels or break the license in more secretive or dangerous ways.

Hope you folks will reconsider what you've done here if you insist on slapping "open" everywhere on all of your material, you should really seriously think about this decision. You may disagree that it is harmful, but know that MANY professionals will always fundamentally disagree with that position, and that at the very least, it "muddies the open source waters".

Cheers.

Interesting...

Not fully open, but the limits are for irrelevant parties. Those companies should pay millions to train their own models if it does not work for them.

@MavenDE

As stated above:
"Keep in mind, I am not arguing your restrictions or the morality behind them, I'm just saying that calling something open when it is not makes the entire free software landscape more confusing."

Sign up or log in to comment