brainsteam.co.uk/brainsteam/content/annotations/2023/01/29/1674990082.md

date	hypothesis-meta	in-reply-to	tags	type	url
2023-01-29T11:01:22	created document flagged group hidden id links permissions tags target text updated uri user user_info 2023-01-29T11:01:22.509728+00:00 title 2301.11305.pdf false __world__ false RDbfNJ_EEe258oPTYQZGNA html incontext json https://hypothes.is/a/RDbfNJ_EEe258oPTYQZGNA https://hyp.is/RDbfNJ_EEe258oPTYQZGNA/arxiv.org/pdf/2301.11305.pdf https://hypothes.is/api/annotations/RDbfNJ_EEe258oPTYQZGNA 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 11561 11236 TextPositionSelector exact prefix suffix type As in prior work, we study a ‘white box’ setting (Gehrmannet al., 2019) in which the detector may evaluate the log prob-ability of a sample log pθ (x). The white box setting doesnot assume access to the model architecture or parameters.While most public APIs for LLMs (such as GPT-3) enablescoring text, some exceptions exist ed samples to perform detection. . While most of our ex-periments TextQuoteSelector https://arxiv.org/pdf/2301.11305.pdf The authors assume white-box access to the log probability of a sample (log p_{\Theta}(x)) but do not require access to the model's actual architecture or weights. 2023-01-29T11:01:22.509728+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/2023/01/29/1674990082

As in prior work, we study a ‘white box’ setting (Gehrmannet al., 2019) in which the detector may evaluate the log prob-ability of a sample log pθ (x). The white box setting doesnot assume access to the model architecture or parameters.While most public APIs for LLMs (such as GPT-3) enablescoring text, some exceptions exist

The authors assume white-box access to the log probability of a sample \(log p_{\Theta}(x)\) but do not require access to the model's actual architecture or weights.