zlacker

[parent] [thread] 2 comments
1. precom+(OP)[view] [source] 2023-05-16 20:05:16
Yep. No page on OpenAI's website about the thousands of underpaid third-world workers that sit and label the data. They will try and build momentum and avoid the "uncomfortable" questions at all costs.
replies(1): >>dahwol+72
2. dahwol+72[view] [source] 2023-05-16 20:16:54
>>precom+(OP)
I empathize with that issue, especially the underpaid part, but superficially that work is still a type of value exchange based on consent: you do the labeling, you get paid (poorly).

Yet for the issue I discussed, there's no value exchange at all. There's no permission or compensation for the people that have done the actual work of producing the training material.

replies(1): >>precom+K6
◧◩
3. precom+K6[view] [source] [discussion] 2023-05-16 20:40:50
>>dahwol+72
Oh yeah. And labeling it as an "AI" further obfuscates it. But apart from small gestures catered to people whose work is very "unique" / identifiable, no one else will get a kickback. They only need to kick the ball further for a couple more years and then it'll become a non-issue as linkrot takes over. Or maybe they use non-public domain stuff, maybe they have secret deals with publishers.

Heck, sometimes even google doesn't pay people for introducing new languages to their translation thingy.

https://restofworld.org/2023/google-translate-sorani-kurdish...

[go to top]