I am confused by the new Open Source AI definition not requiring sharing the training data.
From the official FAQ of the OSI: "This approach would relegate Open Source AI to a niche of AI trainable only on open data [...] That niche would be tiny [...]"
I fail to see the problem here. Wasn't FOSS a small niche before GNU picked up steam?
(Also, if I'm not mistaken, the word should be "trained", not "trainable". A tiny but important difference)
Can someone enlighten me on how this is a sensible idea please? I am open to different opinions on the matter.
It's not.
It's a stance OSI got by its sponsors https://opensource.org/sponsors
Meta in particular was allowed to cancel other teams' votes.
The issues have been reported by several people for months (here a recap https://discuss.opensource.org/t/list-of-unaddressed-issues-of-osaid-rc2/650 ) but OSI ignored them, silenced or censored those people.
See Julia Ferraioli's account of the co-design process: https://www.linkedin.com/posts/juliaferraioli_opensource-opensourceai-osi-activity-7256687721182605313-_wpd
If you feel disgusted as I do, here an initiative to restore (and _maybe_ improve) the OSD through public discussion and IETF rough consensus https://opensourcedefinition.org/wip/
