“Paid collaboration” my ass. More like, let me pay you a pittance and sell your knowledge for 100x.
this is basically phishing for idiots willing to sell their employer for few bucks (and commit a crime in the process), but even so, i have hard time seeing how this could work. they already scrapped all publicly available code in the world. now it is not realistic they would double that amount with this. also there is no guarantee (or even assumption) that the code they would get this way would be inherently of bigger quality than good open source.
so lets say they grow their training data by 10% - can that really make meaningful difference? combined with the fact they would now have to pay some real money for it?
It is possible that publicly available code has now became so much overrun with AI code that they are going for older private repos.
it is, but the question stays - is growing their training dataset by 10% going to make such significant difference that they are willing to actually pay money for it instead of stealing it?