@JonathanBerant

NLP at Tel-Aviv University and Google

cs.tau.ac.il/~joberant/
Joined June 2011
  • Tweets1K
  • Following280
  • Followers3.3K
45Media
Presenting this tomorrow at the 11am poster session at COLM, come say hi!
Jonathan Berant @ COLM 2026@JonathanBerantMar 6
Newish work (arXived in Dec.): Prompts can be ambig, but handling ambig. is context/user dependent. Sometimes the right thing is to ask a clarifying question, sometimes to give multi. answers, and sometimes to just guess. Can we train models to change their strategy per context?
4
19
972
Jonathan Berant @ COLM 2026 retweeted
Max Chen is at COLM@maximillianc_Oct 7
📣 Come check out our work on quantifiably steering model behavior! 📣 We’ll be presenting a poster at 11am on Wednesday. Our team at Google is also hiring! If you are working on long-horizon, personalization, or user models, please DM/visit the Google booth/our poster #COLM
Jonathan Berant @ COLM 2026@JonathanBerantMar 6
Newish work (arXived in Dec.): Prompts can be ambig, but handling ambig. is context/user dependent. Sometimes the right thing is to ask a clarifying question, sometimes to give multi. answers, and sometimes to just guess. Can we train models to change their strategy per context?
1
2
20
1,395
Jonathan Berant @ COLM 2026 retweeted
ACLRollingReview@ReviewAclJun 26
🚨 ARR is recruiting extra emergency reviewers and emergency Area Chairs (ACs) for this cycle. Emergency Reviewer Registration: forms.office.com/Pages/Respo… Emergency AC Registration: forms.office.com/Pages/Respo… Thank you for helping support the ARR review process. #ARR #ACL #NLProc
1
25
3
44
15,935
Jonathan Berant @ COLM 2026 retweeted
Yair Lewis@_lewisyMay 13
We’re hiring our first team members! Well-funded digital health startup building agentic automation for messy, high-friction healthcare ops. Early traction, big problems, real production workflows. Looking for insanely strong product minded SWE. DM me! (Position is in TLV)
4
1
9
617
Jonathan Berant @ COLM 2026 retweeted
Yair Lewis@_lewisyMay 13
אנחנו מגייסים! אוקיי מייעצים לי כאן לכתוב גם בעברית. סטרטאפ צעיר אחרי סיד משמעותי. בונים מערכות אג׳נטיות לhealthcare. מחפשים את המהנדסים הראשונים שלנו: מפתחים עם אוריינטציה מוצרית חזקה, שאוהבים לפתור בעיות קשות, ומתעניינים ביישום אייג׳נטים בפרוד. חושבים שמעניין? דאמו לי ונדבר
1
7
450
Jonathan Berant @ COLM 2026 retweeted
Gal Yona
@_galyoMay 12
It’s 2026 and frontier LLMs STILL hallucinate. Why? In our new ICML 2026 Position Paper, we offer a simple diagnosis and a constructive path forward.
15
38
6
236
13,948
Jonathan Berant @ COLM 2026 retweeted
Max Chen is at COLM@maximillianc_Mar 6
📣Excited to finally share our latest work on quantifiably adapting model behavior based on unique preferences 📣 We teach language models to adjust their clarification behavior using scalar coefficients and find they can generalize to unseen coefficients at inference time!
Jonathan Berant @ COLM 2026@JonathanBerantMar 6
Newish work (arXived in Dec.): Prompts can be ambig, but handling ambig. is context/user dependent. Sometimes the right thing is to ask a clarifying question, sometimes to give multi. answers, and sometimes to just guess. Can we train models to change their strategy per context?
1
14
2,004
Newish work (arXived in Dec.): Prompts can be ambig, but handling ambig. is context/user dependent. Sometimes the right thing is to ask a clarifying question, sometimes to give multi. answers, and sometimes to just guess. Can we train models to change their strategy per context?
1
6
3
36
7,334
more replies
Models also generalize to coefficients that never occurred at training time!
1
246
More generally, training models to respect scalar values that specify a reward in the prompt is useful! There are more results and analyses in the paper, check it out... arxiv.org/abs/2512.04068 With @maximillianc_ @jacobeisenstein @adamjfisch @fantinehuot @rezaa @mlapata

Learning Steerable Clarification Policies with Collaborative Self-play

To handle underspecified or ambiguous queries, AI assistants need a policy for managing their uncertainty to determine (a) when to guess the user intent and answer directly, (b) when to enumerate...

arxiv.org
3
282
Are AI models effective collaborators, or mere assistants awaiting your next command? (arxiv.org/abs/2602.24188) To find out, we make AI collaborate with itself, in private information games: tasks that require sharing private information, like this chess board ordering task.
3
18
2
130
18,285
more replies
AI systems are also overconfident, terminating dialogues long before exhausting their turn budget - even after explicit reminders.
1
10
876
For more discussion, please see the paper! arxiv.org/abs/2602.24188 @jacobeisenstein bravely led this work (but does not tend to post much research here anymore...) Multi-turn collaboration was definitely a key part of this project! @fantinehuot @adamjfisch and @mlapata

MT-PingEval: Evaluating Multi-Turn Collaboration with Private...

We present a scalable and verifiable methodology for evaluating language models in multi-turn interactions, using a suite of collaborative games that require effective communication about private...

arxiv.org
1
14
715
Load more