Sapir Harary @ ACL 2026 @SapirHarary - Twitter Profile

Pinned Tweet

Sapir Harary @ ACL 2026

7 days ago

🚨 Presenting PrefixNLI (Oral) tomorrow at #ACL2026! 🌴 Hallucinations start with a single token. So why wait for a full sentence to detect them? We extend Natural Language Inference to arbitrary text prefixes, detecting factual inconsistencies as soon as they arise and using that signal to guide decoding. Our controlled decoding method delivers major faithfulness gains in summarization. A 3B model matches an 8B model in faithfulness while maintaining similar runtime, using half the memory, and is substantially faster than prior lookahead methods. @SapirHarary, @hirscheran, @lovodkin93, @meetdavidwan, @mohitban47, Ido Dagan 📍 Regatta ⏰ 2:10pm https://t.co/YHmGvbXkMD Hope to see you there! 👋 #NLProc #ACL2026

1

25

7

9

2K

Sapir Harary @ ACL 2026

@SapirHarary

3 days ago

0

1

0

44

Sapir Harary @ ACL 2026

@SapirHarary

3 days ago

It was a pleasure to present our work, PrefixNLI, as an ACL 2026 Oral (top 5% of submissions). Thank you to everyone who attended the talk. And of course, thanks to my co-authors! 📄 Paper: https://t.co/QgoDu1x3Cx #ACL2026 #NLP #AI #MachineLearning #ACL2026NLP #NLProc @aclmeeting

1

17

1

4

1K

Sapir Harary @ ACL 2026

@SapirHarary

7 days ago

#ACL2026NLP @aclmeeting

Sapir Harary @ ACL 2026

@SapirHarary

7 days ago

🚨 Presenting PrefixNLI (Oral) tomorrow at #ACL2026! 🌴 Hallucinations start with a single token. So why wait for a full sentence to detect them? We extend Natural Language Inference to arbitrary text prefixes, detecting factual inconsistencies as soon as they arise and using that signal to guide decoding. Our controlled decoding method delivers major faithfulness gains in summarization. A 3B model matches an 8B model in faithfulness while maintaining similar runtime, using half the memory, and is substantially faster than prior lookahead methods. @SapirHarary, @hirscheran, @lovodkin93, @meetdavidwan, @mohitban47, Ido Dagan 📍 Regatta ⏰ 2:10pm https://t.co/YHmGvbXkMD Hope to see you there! 👋 #NLProc #ACL2026

1

25

7

9

2K

0

3

0

159

Who to follow

shaharglazner

@shaharglazner

Co-Founder & CTO @keepalerting. Likes to think out loud and a long distance runner.

Khal

@Khalil_codeless

👉 https://t.co/CbjN16Iyn7 - A nocode tool to build and monetize APIs! Follow to see me go from $0 to $1bn in ARR 🚀 #nocode #buildinpublic

Natalie Shapira

@NatalieShapira

Tell me about challenges, the unbelievable, the human mind and artificial intelligence, thoughts, social life, family life, science and philosophy.

SapirHarary retweeted

Mohit Bansal

@mohitban47

11 days ago

@aclmeeting @danadaeun @jwlee8877 @duynguyen772 @niu_tianyi @EliasEskin @UTCompSci @icmlconf @cyjustinchen @hyunji_amy_lee @hanqi_xiao @jmin__cho @jhuclsp @jaeh0ng_yoon @NTUsg @ArchikiPrasad @codezakh @gentaiscool @zhan1624 -- PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise (Oral) @SapirHarary , @hirscheran , @lovodkin93 , @meetdavidwan, Ido Dagan @biunlp https://t.co/kKgjYN64oC #ACL2026 Regatta Session 3: Oral Session B: Sun. July 5, 14:00-15:30 PT https://t.co/c8WKST2Lvi

1

15

5

3

701

Sapir Harary @ ACL 2026

@SapirHarary

8 months ago

PrefixNLI introduces fine-grained factuality detection through prefix-level entailment, laying the groundwork for extensions across tasks, domains, and training paradigms. Collab w/ @hirscheran , @lovodkin93 , @meetdavidwan , @mohitban47 , and Ido Dagan @biunlp @unccs @unc_ai_group 📘 Paper: https://t.co/HgrpJnYpBI 💻 Code, datasets, and models: https://t.co/TAXp6j4C6H

0

12

3

2

348

Sapir Harary @ ACL 2026

@SapirHarary

8 months ago

🚨 New paper alert! We’re thrilled to share our new preprint “PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise” ✨ LLMs generate text one token at a time, but factuality checks still wait for a full sentence. We extend NLI to text prefixes, enabling the catching of factual errors as soon as they arise during generation, achieving major faithfulness gains with 25× lower computational cost than lookahead methods. 🧵⬇️

SapirHarary's tweet photo. 🚨 New paper alert!

We’re thrilled to share our new preprint “PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise” ✨

LLMs generate text one token at a time, but factuality checks still wait for a full sentence.

We extend NLI to text prefixes, enabling the catching of factual errors as soon as they arise during generation, achieving major faithfulness gains with 25× lower computational cost than lookahead methods.

🧵⬇️

4

65

18

20

8K

Sapir Harary @ ACL 2026

@SapirHarary

8 months ago

When integrated into decoding, MiniTruePrefixes: ✅ Improves faithfulness of the generated text by +8 faithfulness points ⚡ Runs 25× faster than lookahead methods 💾 A 3B generator + our 1B PrefixNLI model outperforms 8B generator faithfulness at half the memory and no added latency, without retraining