AndrewGrant wrote: ↑Mon Oct 05, 2020 3:36 am
Its a shame to see any project use and abuse the work that went into Leela.
There's a big rant below, beware...
If you fear that your work will be abused, you have other options: 1) not to start, 2) keep the code private, 3) charge for it.
Out of the major deep learning libraries, the only one that's under a GNU license* is OpenNN. Who is still using it?
* Namely, LGPL, the one that obliges devs to allow users to RE the entire code that calls the however small covered lib, to give the users the power to debug mods of the lib instead of simply rolling the breaking changes back and happily using the product as it was
Google has released Tensorflow under Apache 2.0, but that hasn't prevented it from making money off consulting businesses on the reuse of that lib and renting out TPUs, and indirectly off the reputation boost. It's been a win-win. Google has also released a lot of small codebases supplementing its papers (Alpha Zero was just too large and important to be donated). Likewise, the sponsors of MXNet likely don't regret releasing it under that license, and FB doesn't regret opening PyTorch under modified BSD.
Just by making Leela available to Caruana, Ding and the other super-GMs fully for free instead of ad-hominem freemium coaching services, the devs have missed out on a 5 (maybe 6) figure amount. (Well, that was a consequence of Glaurung's GPL, but they could still have done the Fat Fritz cloud trick, for example.)
Btw, you're absolutely right not to release the training code for the Ethereal nets.
And fwiw, SF isn't a villain, beating it didn't save the humanity
Leela just improved the SOTA a little and gave us a different flavor of chess (the latter had also been done by Fizbo, iCE, Wasp, Rodent et al. without beating SF). Even Komodo isn't a villain, it's affordable and would provide decent chess knowledge even if SF didn't exist.