Embed Notice

HTML Code

<blockquote style="position: relative; padding-left: 55px;"><section><a href="https://friend.camp/users/aparrish/statuses/110946172910597391">allison (aparrish@friend.camp)'s status on Friday, 25-Aug-2023 04:15:23 JST</a><a href="https://friend.camp/@aparrish" title="aparrish@friend.camp"><img src="https://gnusocial.jp/avatar/16561-48-20240109180650.webp" width="48" height="48" alt="allison" style="position: absolute; left: 0; top: 0;">allison</a></section><article><p>i am going to die from laughter. the entire appeal of LLMs to capitalism is its potential for obviating the cost of creative labor. it would be hilarious if the only way to prevent LLMs from eating themselves is... to pay people for their creative labor</p></article><footer><a rel="bookmark" href="https://gnusocial.jp/conversation/1985372#notice-3900216">In conversation</a><time datetime="2023-08-25T04:15:23+09:00" title="Friday, 25-Aug-2023 04:15:23 JST">Friday, 25-Aug-2023 04:15:23 JST</time> <span>from <span><a href="https://friend.camp/@aparrish/110946172910597391" rel="external" title="Sent from friend.camp via ActivityPub">friend.camp</a></span></span><a href="https://friend.camp/@aparrish/110946172910597391">permalink</a><h4>Attachments</h4><ol><li><label><a rel="external" href="https://gnusocial.jp/attachment/1447711">Another potential solution, at least for more general-task generative models, is to pay human creators for making new work, based on the intrinsic value of the work relative to the field in which it was created (in other words: don’t pay people as if they were mechanical turks creating content for the sake of creating novel training data.) Pay people within their fields, as judged by the standards of those fields, and there will ostensibly continue to be novel content available for training. This will not exactly ensure there is always a pipeline of new text, new images, and new audio that will naturally keep pace with cultural change and bring beneficial diversity with it. There could still be stereotypes and discrimination reducing the type of content that is valued and there could be barriers that reduce the big model makers from training on the highly valued content. But paying people for their contributions, as judged by the standards of their fields, will certainly help. Unfortunately, the pay rates for journalism, tv, movies, music, recipes/food production, and podcasts have been declining as those fields hit the platform economy and open-source content creators may begin to object to having their work hoovered up by tech companies.</a></label><br><a href="https://friend.camp/system/media_attachments/files/110/946/156/666/563/490/original/af2be3a0e8c3c52b.png" rel="external">https://friend.camp/system/media_attachments/files/110/946/156/666/563/490/original/af2be3a0e8c3c52b.png</a></li></ol></footer></blockquote>

Corresponding Notice

Embed this notice
allison (aparrish@friend.camp)'s status on Friday, 25-Aug-2023 04:15:23 JST allison
i am going to die from laughter. the entire appeal of LLMs to capitalism is its potential for obviating the cost of creative labor. it would be hilarious if the only way to prevent LLMs from eating themselves is... to pay people for their creative labor
In conversationFriday, 25-Aug-2023 04:15:23 JST from friend.camppermalink
Attachments
1. Another potential solution, at least for more general-task generative models, is to pay human creators for making new work, based on the intrinsic value of the work relative to the field in which it was created (in other words: don’t pay people as if they were mechanical turks creating content for the sake of creating novel training data.) Pay people within their fields, as judged by the standards of those fields, and there will ostensibly continue to be novel content available for training. This will not exactly ensure there is always a pipeline of new text, new images, and new audio that will naturally keep pace with cultural change and bring beneficial diversity with it. There could still be stereotypes and discrimination reducing the type of content that is valued and there could be barriers that reduce the big model makers from training on the highly valued content. But paying people for their contributions, as judged by the standards of their fields, will certainly help. Unfortunately, the pay rates for journalism, tv, movies, music, recipes/food production, and podcasts have been declining as those fields hit the platform economy and open-source content creators may begin to object to having their work hoovered up by tech companies.
  https://friend.camp/system/media_attachments/files/110/946/156/666/563/490/original/af2be3a0e8c3c52b.png

Public

Embed Notice

HTML Code

Corresponding Notice