Hacker News Reader

Comment by OisinMoran | Hacker News Reader

Twitter thread from the author of the original algorithm: https://twitter.com/DrBeef_/status/1055285640420483073 (worth reading the whole thread, the art collective reply)

GitHub link: https://github.com/robbiebarrat/art-dcgan The README was last edited 6 months ago and contains the bolded text: "When using any outputs of the models, credit me. Don't sell the outputs of the pre-trained models, modified or not. If you have any questions email me before doing anything."

View on HN

Oh gosh, he even links an issue where one of the guys who sold the painting (Hugo Caselles-Dupre, hence the username Caselles) literally asks for help setting up.

So Hugo asks for help with setup because he can't get it to work, and Robbie (original author) replies "working on a rewrite of the codebase, should be soon" while Hugo keeps pushing for updates. Turns out it's trickier than expected so Robbie says "sorry, might take longer". Hugo's response?

> "Ok, really disappointed. Could you at least provide the code for the 128x128 Gan in Keras ?"

He's literally sitting there for months pushing for poor Robbie to do the free work for him so that he can try to profit from it.

Maybe they really did make a lot of changes to the model. But from the conversation in that issue thread and the striking similarities between the painting and the other ones that Robbie had made over a year ago, it doesn't feel likely to me that that's the case.

Hopefully some justice will come out of this situation.

As an aside, according to his bio Robbie's only 19 (and already doing this level of research!). Sad to see younger researchers being taken advantage of like this.

https://github.com/robbiebarrat/art-DCGAN/issues/3#event-140...

It's hard to see where anyone is being 'taken advantage' of here. If you put your code on github with an open license, you are letting people to have their way with it. This is like if Tim Berners Lee were mad that Google 'stole' his internet and made a lot of money off it.

Also the DCGAN architecture is from Chintala et al, and that itself is based on work by Goodfellow et al. Where does this end? Does Robbie Barrat 'own' the DCGAN NN architecture now?

I think the "taken advantage of" thing is more about the crappy way they behaved on the Github issues ("really disappointing") etc than the idea of using his code.

Of the whole thing, this is what struck me as distasteful. Manipulative and degrading.

The French kids owe the Stanford kid like half to all of it.

The French "kids" are 25y afaik.

Average peoples sense of entitlement when it comes to exploiting others freely given work probably comes as no surprise to anyone who has ever put anything open source up on Github or anywhere else.

> If you put your code on github with an open license, you are letting people to have their way with it.

And enough people don't understand this for it to be a problem.

Edit: Removed bad phrasing.

> In fact, making your code free for everyone to use probably is hurting the web,

It's the main reason there is a widely used public web rather than it, if it succeeded at all, being another academic niche tool alongside gopher.

Well, assuming that's true, it has little-to-no impact on how things should be going forward. The web is very clearly here to stay. At the very least people who want credit and payment for their work should not be encouraged to use OSI licenses.

Licenses aren't retroactive. If you don't want people making money off your stuff (yes, even with direct copy-and-pastes), use CC-BY-NC or similar, not the BSD License.

I feel like this screencap posted by Obvious was the most illuminating part of that twitstorm: https://pbs.twimg.com/media/DqVdWJSW4AAOtXD?format=jpg

The truth comes in as a quick afterthought, "and of course to finance our research on art and ML".

The whole discussion is meant to elicit "a pass". Dishonest.

If Obvious hadn't been successful with their goals, Robbie would not (and did not) care a bit that his work was being used. It only became contentious when they were able to raise a lot of money. It's funny.

Well, obviously. Robbie said in his Github readme: "When using any outputs of the models, credit me. Don't sell the outputs of the pre-trained models, modified or not. If you have any questions email me before doing anything."

I feel for the guy, but keep in mind: just putting some text in your readme does not create a legally binding contract.

Reminds me of Infiniminer vs Minecraft, or GNU vs Linux. Or, especially, the Box2D creator asking Rovio to credit Box2D in Angry Birds. (Did they ever do so?) It sounds like Obvious at least attempted to give credit, but were ignored by the media.

https://www.geek.com/games/box2d-creator-asks-rovio-for-angr...

Interesting. Considering the readme is the first thing the person sees on the github link and nothing about the license itself precludes other contracts, could the bolded text in this context be considered a verbal contract? Or at least enforceable as such?

Edit: Actually, there was a change to the license surrounding pre-trained models and sales back in June, so it's even baked into an obvious contract. A lot of the licensing conversation here is somewhat mooted by this.

https://github.com/robbiebarrat/art-DCGAN/blob/master/LICENS...

Obvious references "the collection", which presumably includes the piece auctioned at Christies, in a DM to Robbie in April, before that clause was added: https://twitter.com/obv_ious/status/1055352403837497344

The interesting thing is that you cannot copyright computer-generated images.

I guess open sourcing things is less fun when someone profits off your work. DrBeef seems to have learned this the hard way.

Reply on news.ycombinator.com