I expected better from Google
77 points by FedericoSchonborn
77 points by FedericoSchonborn
Good to know stripping attribution from open source libraries is still alive and kicking in the age of clankers.
Isn’t that literally what they’re for? Sucking up the world’s open source code and reproducing it without attribution, that’s literally what they’ve been trained to do.
I guess there's a subtle difference between "I reproduced this thing from memory that looks uncannily like some other work I \"read\" some time ago", and good old, "I copy/pasted and deleted your name, soz" which has been around as long as people have been releasing code to the world.
Truly some "surely leopards won't eat my face" nonsense. There is no honor among thieves; you shouldn't expect other slopmongers not to plagiarize you.
This is not a slop gone wrong story though. They had the code taken directly and on purpose and even keeping the names, and then later removed the names on purpose.
My point is that the same moral bankruptcy is demonstrated in slicing and dicing the entirety of human culture to make slop machines and in scratching authors' names off of open source libraries.
The irony seems entirely lost considering even this blogpost is apparently not worth the effort to hand-write.
It confuses me all that "disappointment" in the post.
Did Google comply the licence or not? I admit I scanned through the post, so may be it is there and I missed it.
Edit: the "original" is under the Apache licence, but I'm not sure the licence was applied correctly as the files cited as example didn't have the licence boilerplate in the header.
https://www.apache.org/foundation/license-faq.html#Apply-My-Software
Although is not a requirement. Could be the files were distributed independently from the licence file. I guess :D
Did they comply? Not. That's what the post is about.
Then why all the disappointment? Google don't have permission to use the code because they didn't comply with the licence.
Even assuming the licence was not applied correctly, that makes the effective licence more restrictive by default, not less. The Apache licence is what allowed Google to use this code in the first place.
Yeah sure lets just accept this race to the bottom
I loathe this mindset. "You shouldn't surprised that companies act unethically" is not the same statement as "It's ok for companies to act unethically". Observation is not acceptance!
In fact it's more unreasonable and harmful to expect naively ethical behaviour from corporations, in contravention of all the evidence to the contrary. It's only by treating them with the suspicion and hostility that they deserve, that the "race to the bottom" can be avoided!
One can disagree with someone's behavior without being surprised by it. What Google did here is egregious -- but is it more egregious than what they already did to train Gemini?
Yes, it's more egregious than what they did to train Gemini.
Why? Both are, at least to me, clear violations of both the license text of ooen source projects, and the social contract of open source.
Are they at least violations of different magnitude?
Here's a personal anecdote. I am happy for my open source code to end up in training datasets, because I later benefit from the models being influenced, however slightly, by my personal style. I also believe that falls under "fair use" terms in US law, and I'm OK with that.
I'm unhappy if someone copies my code directly into their project without crediting me. In addition to that feeling scummy it's also a legal violation of the terms of my licenses.
Are they at least violations of different magnitude?
Yeah, absolutely. Doing it lots of times to tens of thousands of authors is way worse than doing it to a few authors once.
I am happy for my open source code to end up in training datasets
So you got what you wanted. That's great! But shouldn't other people have a choice?
I don't like that people who don't want their stuff in training data have their stuff in training data.
The best solution we've seen for that right now is opting out, but forcing people to opt out - especially when, once their stuff is in the training data it can't be removed - is a clearly absurd solution.
So, given that I want this technology to exist because I find it useful and think it delivers enough benefit to humanity to be worth the training cost, I find myself paraphrasing John Cena in Pluribus:
Given our druthers, would we choose [to scrape the web for training data]? [...] Honestly, we're not that keen on it ourselves. But we're left with little choice.
Exactly. People are completely missing the fact that the one complaining is heavily developing by and for AI automation. There’s so many level of irony here (including that the post is either generated, either badly written)
I was about to say that this is why license selection matters, but as I read the Apache license, which was what the project was licensed under has the following conditions:
In this case You is Google.
4b: You must cause any modified files to carry prominent notices stating that You changed the files.
4c: You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works
4d: If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding those notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear.
So, from an inspection of both repos, it looks like Google has illegally copied the software.
Remember, Open Source/Free Software is not free from all responsibility.
Did you really expect better from Google? Does anyone really believe that Google isn't prioritizing profit over literally everything else?
This doesn't surprise me in the slightest.
The company that actively dropped the motto "don't be evil"...
Even with this, though, I still believe Google to do the right thing (even if it is not using the said OSS and just doing their own) a lot more then any other big tech name.
I hope this is an exception, and not a rule.
I'd recommend you update your assumptions, in that case. :-/ Google does almost nothing for anyone except Google.
"On the bright side, the version powering Minitap today is closed source."
Well, at least it has a happy ending! :D
/sarcasm
Disclaimer: I work for Google and I used to work for Google's Open Source office, but I am not speaking for either.
It does appear the right thing was done here in a follow up commit: https://github.com/google/artemis/commit/371aa6df56880643da57b30da936e9812fb0ec66
I will say that Google OSS takes licensing extremely seriously, if only because it's the right thing to do. You can see much of the internal documentation published externally, e.g. Google's licensing policy.
That commit September 17th. The linked blog posts September 11th - so pretty clear the commit was in response to the complaint.
I think the blog post should update to reflect this though. The fact that the issue was remedied is material.
Oh, so they added license notices after they were publicly called out? Yeah, that really demonstrates a commitment to doing the right thing for its own sake 🙄
Given that originally they even put in a private email address (outlook.com while at Google, okay), and that their github profile doesn't look very accomplished, that person probably missed their "open source contributions" training: The publication guidelines ask to "Please associate your commit with your google.com email unless: You have a history of contributing to the repo under a different email before your employment at Google (or something about hard to submit via google.com email)"
Since this is now fixed (the current version of the file published with something looking like a real name, and a corp email address), my assumption is that OSPO reacted once they saw the mess that is this repository, and gave that person a quick intro on the half-dozen things they got wrong while working with the code, and how to fix them.
(Disclosure: I used to work at Google, with lots of Open Source interactions, including some that I brought with me to the company, so I'm quite familiar with that part of the policy.)
I no longer work for Google, but worked for OSPO as well, and, yeah. I am sure that at least OSPO (or what remains of it) is having a lot of very angry emails with someone on the Artemis team.
Goodness, how do you even evaluate the liability for something like this these days?
It would surely be unfair to ask for 1% of 1 year of Google's revenues, even though Android and AI make up two substantial verticals. Maybe one ten thousandth part?
It doesn't look much like deterance, but I guess the alternative is we could acknowledge that intellectual property is dead and just let loose.
I guess the alternative is we could acknowledge that intellectual property is dead and just let loose.
Isn't that just what LLMs are doing as-is?
They can't have cared that much -- they didn't even bother to write about it themselves. And, what Google did seems to be legal (if not polite), at least in many jurisdictions: AI-generated code isn't generally considered copyrightable.