brainsteam.co.uk/brainsteam/content/annotations/2023/01/29/1674989272.md

date	hypothesis-meta	in-reply-to	tags	type	url
2023-01-29T10:47:52	created document flagged group hidden id links permissions tags target text updated uri user user_info 2023-01-29T10:47:52.638117+00:00 title 2301.11305.pdf false __world__ false YYCjTJ_CEe2thKdVH4ie0A html incontext json https://hypothes.is/a/YYCjTJ_CEe2thKdVH4ie0A https://hyp.is/YYCjTJ_CEe2thKdVH4ie0A/arxiv.org/pdf/2301.11305.pdf https://hypothes.is/api/annotations/YYCjTJ_CEe2thKdVH4ie0A 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 30225 30062 TextPositionSelector exact prefix suffix type We simulate human re-vision by replacing 5 word spans of the text with samplesfrom T5-3B until r% of the text has been replaced, andreport performance as r varies. n increasingly heavily revised. Figure 5 shows that De-tectGPT TextQuoteSelector https://arxiv.org/pdf/2301.11305.pdf I question the trustworthiness of this simulation - human edits are probably going to be more sporadic and random. 2023-01-29T10:47:52.638117+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/2023/01/29/1674989272

We simulate human re-vision by replacing 5 word spans of the text with samplesfrom T5-3B until r% of the text has been replaced, andreport performance as r varies.

I question the trustworthiness of this simulation - human edits are probably going to be more sporadic and random.