Rom Himelstein

According to our database, Rom Himelstein authored at least 7 papers between 2025 and 2026.

Collaborative distances:

• Dijkstra number of five.
• Erdős number of four.

2026

Step-Wise Refusal Dynamics in Autoregressive and Diffusion Language Models.

Eliron Rahimi, Elad Hirshel, Rom Himelstein, Amit LeVi, Avi Mendelson, Chaim Baskin

CoRR, February, 2026

Silenced Biases: The Dark Side LLMs Learned to Refuse.

Rom Himelstein, Amit LeVi, Brit Youngmann, Yaniv Nemcovsky, Avi Mendelson

Proceedings of the Fortieth AAAI Conference on Artificial Intelligence, 2026

2025

You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations.

Amit LeVi, Raz Lapid, Rom Himelstein, Yaniv Nemcovsky, Ravid Shwartz-Ziv, Avi Mendelson

CoRR, November, 2025

Silent Tokens, Loud Effects: Padding in LLMs.

Rom Himelstein, Amit LeVi, Yonatan Belinkov, Avi Mendelson

CoRR, October, 2025

Leveraging NTPs for Efficient Hallucination Detection in VLMs.

Ofir Azachi, Kfir Eliyahu, Eyal El Ani, Rom Himelstein, Roi Reichart, Yuval Pinter, Nitay Calderon

CoRR, September, 2025

Enhancing Jailbreak Attacks via Compliance-Refusal-Based Initialization.

Amit LeVi, Rom Himelstein, Yaniv Nemcovsky, Avi Mendelson, Chaim Baskin

CoRR, February, 2025

Jailbreak Attack Initializations as Extractors of Compliance Directions.

Amit LeVi, Rom Himelstein, Yaniv Nemcovsky, Avi Mendelson, Chaim Baskin

Proceedings of the Findings of the Association for Computational Linguistics: EMNLP 2025, 2025