These companies were already training their models on Wikipedia’s wealth of information anyway. In this way Wikipedia is earning some revenue from the thing that was already happening, letting them put that money back into the non-profit.
Yeah I mean Wikipedia has regular dump files where you can just… download its entire content, or parts of it if you so wish. Getting money instead for that bandwidth is immediately an improvement
Wikipedia is also public knowledge. I personally think it’s OK to use for training data.
However, there are other concerns about inaccuracies and some info on the site needs to be scrutinized and verified just because anyone can edit it, and the LLMs getting trained can’t do that part.
What actually bothers me about this is that the companies training the LLMs are going to put them behind paywalls, removing the public knowledge part of this.
Let me be the devil’s advocate for this one.
These companies were already training their models on Wikipedia’s wealth of information anyway. In this way Wikipedia is earning some revenue from the thing that was already happening, letting them put that money back into the non-profit.
Yeah I mean Wikipedia has regular dump files where you can just… download its entire content, or parts of it if you so wish. Getting money instead for that bandwidth is immediately an improvement
They weren’t using those dumps. They were scraping the main site, at incredible expense to Wikipedia.
Wikipedia is also public knowledge. I personally think it’s OK to use for training data.
However, there are other concerns about inaccuracies and some info on the site needs to be scrutinized and verified just because anyone can edit it, and the LLMs getting trained can’t do that part.
What actually bothers me about this is that the companies training the LLMs are going to put them behind paywalls, removing the public knowledge part of this.