Re:Ask HN: What can I personally do to fight AGAINST the Cop...
by visarga on Friday November 04, 2022 @02:07PM
(#33471275)
The dataset is so large they only do one single pass, and the content is deduplicated. A study showed that it's usually duplicated content that is overfitted and later regurgitated. Examples that are repeated in many places might have slight variations that help them pass the filter.
[ Parent
| on HN ]
Re:Ask HN: What can I personally do to fight AGAINST the Cop...
by Kim_Bruning on Friday November 04, 2022 @03:32PM
(#33472481)
Interesting!
Do you have a link(s) to the study, or more details on training?
Re:Ask HN: What can I personally do to fight AGAINST the Cop...
by visarga on Friday November 04, 2022 @04:08PM
(#33473032)
Deduplicating Training Data Makes Language Models Better
https://arxiv.org/pdf/2107.06499
Quantifying Memorization Across Neural Language Models