brainsteam.co.uk/brainsteam/content/annotations/2022/11/23/1669236617.md

date	hypothesis-meta	in-reply-to	tags	type	url
2022-11-23T20:50:17	created document flagged group hidden id links permissions tags target text updated uri user user_info 2022-11-23T20:50:17.668925+00:00 title 2022.naacl-main.167.pdf false __world__ false b_EbpGtwEe2m8tfhSKM2EQ html incontext json https://hypothes.is/a/b_EbpGtwEe2m8tfhSKM2EQ https://hyp.is/b_EbpGtwEe2m8tfhSKM2EQ/aclanthology.org/2022.naacl-main.167.pdf https://hypothes.is/api/annotations/b_EbpGtwEe2m8tfhSKM2EQ admin delete read update acct:ravenscroftj@hypothes.is acct:ravenscroftj@hypothes.is group:__world__ acct:ravenscroftj@hypothes.is prompt-models NLProc selector source end start type 2221 1677 TextPositionSelector exact prefix suffix type Suppose a human is given two sentences: “Noweapons of mass destruction found in Iraq yet.”and “Weapons of mass destruction found in Iraq.”They are then asked to respond 0 or 1 and receive areward if they are correct. In this setup, they wouldlikely need a large number of trials and errors be-fore figuring out what they are really being re-warded to do. This setup is akin to the pretrain-and-fine-tune setup which has dominated NLP in recentyears, in which models are asked to classify a sen-tence representation (e.g., a CLS token) into some task instructions.1 Introduction ∗Unabridged version available on TextQuoteSelector https://aclanthology.org/2022.naacl-main.167.pdf This is a really excellent illustration of the difference in paradigm between "normal" text model fine tuning and prompt-based modelling 2022-11-23T20:50:17.668925+00:00 https://aclanthology.org/2022.naacl-main.167.pdf acct:ravenscroftj@hypothes.is display_name James Ravenscroft	https://aclanthology.org/2022.naacl-main.167.pdf	prompt-models NLProc hypothesis	annotation	/annotation/2022/11/23/1669236617

Suppose a human is given two sentences: “Noweapons of mass destruction found in Iraq yet.”and “Weapons of mass destruction found in Iraq.”They are then asked to respond 0 or 1 and receive areward if they are correct. In this setup, they wouldlikely need a large number of trials and errors be-fore figuring out what they are really being re-warded to do. This setup is akin to the pretrain-and-fine-tune setup which has dominated NLP in recentyears, in which models are asked to classify a sen-tence representation (e.g., a CLS token) into some

This is a really excellent illustration of the difference in paradigm between "normal" text model fine tuning and prompt-based modelling