brainsteam.co.uk/brainsteam/content/annotations/2023/01/29/1674989844.md

date	hypothesis-meta	in-reply-to	tags	type	url
2023-01-29T10:57:24	created document flagged group hidden id links permissions tags target text updated uri user user_info 2023-01-29T10:57:24.658922+00:00 title 2301.11305.pdf false __world__ false tnNraJ_DEe2YBceDAVt0Uw html incontext json https://hypothes.is/a/tnNraJ_DEe2YBceDAVt0Uw https://hyp.is/tnNraJ_DEe2YBceDAVt0Uw/arxiv.org/pdf/2301.11305.pdf https://hypothes.is/api/annotations/tnNraJ_DEe2YBceDAVt0Uw admin delete read update acct:ravenscroftj@hypothes.is acct:ravenscroftj@hypothes.is group:__world__ acct:ravenscroftj@hypothes.is chatgpt detecting gpt selector source end start type 22349 22098 TextPositionSelector exact prefix suffix type Empirically, we find predictive entropy to be positively cor-related with passage fake-ness more often that not; there-fore, this baseline uses high average entropy in the model’spredictive distribution as a signal that a passage is machine-generated. tropy) predictive distributions. While our main focus is on zero TextQuoteSelector https://arxiv.org/pdf/2301.11305.pdf this makes sense and aligns with the [gltr](http://gltr.io) - humans add more entropy to sentences by making unusual choices in vocabulary that a model would not. 2023-01-29T10:57:24.658922+00:00 https://arxiv.org/pdf/2301.11305.pdf acct:ravenscroftj@hypothes.is display_name James Ravenscroft	https://arxiv.org/pdf/2301.11305.pdf	chatgpt detecting gpt hypothesis	annotation	/annotations/2023/01/29/1674989844

Empirically, we find predictive entropy to be positively cor-related with passage fake-ness more often that not; there-fore, this baseline uses high average entropy in the model’spredictive distribution as a signal that a passage is machine-generated.

this makes sense and aligns with the [gltr](http://gltr.io) - humans add more entropy to sentences by making unusual choices in vocabulary that a model would not.