Amir David Nissan cohen (@amirdnc) 's Twitter Profile
Amir David Nissan cohen

@amirdnc

ID: 1481572465692315654

calendar_today13-01-2022 10:24:57

11 Tweet

58 Followers

37 Following

(((ู„()(ู„() 'yoav))))๐Ÿ‘พ (@yoavgo) 's Twitter Profile Photo

In this work, Amir (and Shauli Ravfogel, and another author who cannot reveal themselves before paper acceptance) shows that diversity is much more important than quantity for model training. u.cs.biu.ac.il/~yogo/diverse-โ€ฆ

In this work, Amir (and <a href="/ravfogel/">Shauli Ravfogel</a>, and another author who cannot reveal themselves before paper acceptance) shows that diversity is much more important than quantity for model training. 

u.cs.biu.ac.il/~yogo/diverse-โ€ฆ
Avi Caciularu (@clu_avi) 's Twitter Profile Photo

Meet ๐Ÿ˜๏ธQAmden๐Ÿ˜๏ธ your new effective general multi-document model๐Ÿ“„๐Ÿ“„๐Ÿ“„ In our #ACL2023 paper, we propose a new and effective QA-based multi-document pre-training method, specializing in multi-document inputs! w/ Matthew Peters Jacob Goldberger Ido Dagan Arman Cohan 1/n

Meet ๐Ÿ˜๏ธQAmden๐Ÿ˜๏ธ your new effective general multi-document model๐Ÿ“„๐Ÿ“„๐Ÿ“„

In our #ACL2023 paper, we propose a new and effective QA-based multi-document pre-training method, specializing in multi-document inputs!

w/ <a href="/mattthemathman/">Matthew Peters</a> <a href="/JacobGoldberge1/">Jacob Goldberger</a> Ido Dagan <a href="/armancohan/">Arman Cohan</a> 

1/n
Royi Rassin (@royirassin) 's Twitter Profile Photo

๐Ÿ“ข New work ๐Ÿš€ With Eran Hirsch @ACL2025 ๐Ÿ‡ฆ๐Ÿ‡น Daniel Glickman Shauli Ravfogel (((ู„()(ู„() 'yoav))))๐Ÿ‘พ Gal Chechik ๐Ÿ”—Demo:huggingface.co/spaces/Royir/Sโ€ฆ ๐Ÿ”—Preprint:arxiv.org/abs/2306.08877 We tackle improper binding in diffusion models, where the linguistic structure in the prompt is not reflected in the generation.

๐Ÿ“ข New work ๐Ÿš€

With <a href="/hirscheran/">Eran Hirsch @ACL2025 ๐Ÿ‡ฆ๐Ÿ‡น</a> Daniel Glickman <a href="/ravfogel/">Shauli Ravfogel</a> <a href="/yoavgo/">(((ู„()(ู„() 'yoav))))๐Ÿ‘พ</a> <a href="/GalChechik/">Gal Chechik</a>

๐Ÿ”—Demo:huggingface.co/spaces/Royir/Sโ€ฆ
๐Ÿ”—Preprint:arxiv.org/abs/2306.08877

We tackle improper binding in diffusion models, where the linguistic structure in the prompt is not reflected in the generation.
Uri Katz (@urikauri) 's Twitter Profile Photo

We are excited to introduce our EMNLP 2023 findings paper - โ€œNERetrieve: Dataset for Next Generation Named Entity Recognition and Retrievalโ€ arxiv.org/abs/2310.14282 github.com/katzurik/NERetโ€ฆ Joint w/ Matan Vetzler ,Amir David Nissan cohen and (((ู„()(ู„() 'yoav))))๐Ÿ‘พ

We are excited to introduce our EMNLP 2023 findings paper - โ€œNERetrieve: Dataset for Next Generation Named Entity Recognition and Retrievalโ€
arxiv.org/abs/2310.14282
github.com/katzurik/NERetโ€ฆ
Joint w/ <a href="/vetzler78091/">Matan Vetzler</a>  ,<a href="/AmirDNC/">Amir David Nissan cohen</a>  and <a href="/yoavgo/">(((ู„()(ู„() 'yoav))))๐Ÿ‘พ</a>
Amir David Nissan cohen (@amirdnc) 's Twitter Profile Photo

ื›ืžื• ื›ืœ ืžื•ื“ืœ ื—ื“ืฉ ื‘ืขื‘ืจื™ืช Shaltiel ื•ืื ื™ ื”ืจืฆื ื• ื›ืžื” ื‘ื“ื™ืงื•ืช ืขืœ ื“ืื˜ื” ืกื˜ื™ื ืงื™ื™ืžื™ื ื‘ืขื‘ืจื™ืช. ื•ื”ืชื•ืฆืื•ืช... ืงืฆืช ืžื•ื–ืจื•ืช. ื”ื‘ื“ื™ืงื•ืช ื”ืŸ: ื•ื™ื ื•ื’ืจื“ - ื“ืื˜ื” ืกื˜ ืฉืœ common sense HeQ - ื“ืื˜ื” ืกื˜ ืฉืœ ืฉืืœื•ืช ื•ืชืฉื•ื‘ื•ืช ืกื ื˜ื™ืžื ื˜ ืื ื—ื ื• ืจื•ืื™ื ื™ืจื™ื“ื” ืžืฉืžืขื•ืชื™ืช ื‘ืชื•ืฆืื•ืช ื‘ื™ื—ืก ืœืžื•ื“ืœ ื”ืžืงื•ืจื™ ืฉืœ ื’ื•ื’ืœ. ื”ืžื•ื“ืœ ืขื•ื‘ื“ ืœื›ื?

ื›ืžื• ื›ืœ ืžื•ื“ืœ ื—ื“ืฉ ื‘ืขื‘ืจื™ืช <a href="/SShmidman/">Shaltiel</a> ื•ืื ื™ ื”ืจืฆื ื• ื›ืžื” ื‘ื“ื™ืงื•ืช ืขืœ ื“ืื˜ื” ืกื˜ื™ื ืงื™ื™ืžื™ื ื‘ืขื‘ืจื™ืช. ื•ื”ืชื•ืฆืื•ืช... ืงืฆืช ืžื•ื–ืจื•ืช.  ื”ื‘ื“ื™ืงื•ืช ื”ืŸ:

ื•ื™ื ื•ื’ืจื“ - ื“ืื˜ื” ืกื˜ ืฉืœ common sense
HeQ - ื“ืื˜ื” ืกื˜ ืฉืœ ืฉืืœื•ืช ื•ืชืฉื•ื‘ื•ืช
 ืกื ื˜ื™ืžื ื˜ 

ืื ื—ื ื• ืจื•ืื™ื ื™ืจื™ื“ื” ืžืฉืžืขื•ืชื™ืช ื‘ืชื•ืฆืื•ืช ื‘ื™ื—ืก ืœืžื•ื“ืœ ื”ืžืงื•ืจื™ ืฉืœ ื’ื•ื’ืœ. ื”ืžื•ื“ืœ ืขื•ื‘ื“ ืœื›ื?
Shaltiel (@sshmidman) 's Twitter Profile Photo

ืชื•ื“ื” ืœื™ื! ื’ื ื›ืืŸ ืื ื™ ื•Amir David Nissan cohen ื”ืจืฆื ื• ื‘ื“ื™ืงื•ืช ืขืœ ื”ืžื•ื“ืœ ืขืœ ื“ืื˜ืกื˜ื™ื ืงื™ื™ืžื™ื ื‘ืขื‘ืจื™ืช, ื•ืขื“ื™ื™ืŸ ื”ืชื•ืฆืื•ืช ืงืฆืช ืžื•ื–ืจื•ืช: ืคืจื˜ื™ื ื˜ื›ื ื™ื™ื: V1 ื”ื•ืจืฅ ืขื: 1. ื’ื™ืจืกืช ื˜ืจื ืกืคื•ืจืžืจื– 4.38.1 2. ื“ื™ื•ืง ืžืœื (ื‘ืœื™ ืงื•ื•ื ื˜ื™ื–ืฆื™ื”) 3. ื‘ืœื™ BOS V2 ื”ื•ืจืฅ ืขื: 1. ื’ื™ืจืกืช ื˜ืจื ืกืคื•ืจืžืจื– 4.38.2 2. ื“ื™ื•ืง ืžืœื (ื‘ืœื™ ืงื•ื•ื ื˜ื™ื–ืฆื™ื”) 3. ืขื BOS

ืชื•ื“ื” ืœื™ื! ื’ื ื›ืืŸ ืื ื™ ื•<a href="/AmirDNC/">Amir David Nissan cohen</a>  ื”ืจืฆื ื• ื‘ื“ื™ืงื•ืช ืขืœ ื”ืžื•ื“ืœ ืขืœ ื“ืื˜ืกื˜ื™ื ืงื™ื™ืžื™ื ื‘ืขื‘ืจื™ืช, ื•ืขื“ื™ื™ืŸ ื”ืชื•ืฆืื•ืช ืงืฆืช ืžื•ื–ืจื•ืช:

ืคืจื˜ื™ื ื˜ื›ื ื™ื™ื:
V1 ื”ื•ืจืฅ ืขื:
1. ื’ื™ืจืกืช ื˜ืจื ืกืคื•ืจืžืจื– 4.38.1
2. ื“ื™ื•ืง ืžืœื (ื‘ืœื™ ืงื•ื•ื ื˜ื™ื–ืฆื™ื”)
3. ื‘ืœื™ BOS
V2 ื”ื•ืจืฅ ืขื:
1. ื’ื™ืจืกืช ื˜ืจื ืกืคื•ืจืžืจื– 4.38.2
2. ื“ื™ื•ืง ืžืœื (ื‘ืœื™ ืงื•ื•ื ื˜ื™ื–ืฆื™ื”)
3. ืขื BOS
Amir David Nissan cohen (@amirdnc) 's Twitter Profile Photo

Led by Aviya Maimon, our new paper redefines how we evaluate LLMs. Instead of one flat leaderboard score, we uncover the latent skillsโ€”reasoning, comprehension, ethics, precision & moreโ€”that really shape LLM ability. Think: psychometrics meets AI. link: arxiv.org/pdf/2507.20208