13 papers
Generalization Is Stability, Not Accuracy: Multi-Axis Evaluation of LLMs
Nagham Omar, Mahmoud Jabarin, Maya Rozenshtein +3
Generalization in large language models (LLMs) is the ability to produce consistent and semantically stable outputs when the same input is expressed in different ways. Existing wor…
Latent Undertow: How Ordinary Typos Break Probes
Elad David, Max Fomin, Amit LeVi
LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model's response substantively unchanged. Yet probes that detect malic…
Linguistic Firewall: Geometry as Defense in Multi-Agent Systems Routing
Dvir Alsheich, Adar Peleg, Ben Hagag +3
The rapid integration of Large Language Models (LLMs) has driven the evolution of Multi-Agent Systems (MAS), where specialized agents collaborate to execute complex workflows. Effe…
Unsupervised Features Mining via Activation Geometry
Amit LeVi, Elad David, Max Fomin
Interpretability methods aim to reveal the features represented inside large language models (LLMs). Many existing methods begin with labeled examples of a human-defined concept th…
Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring
Max Fomin, Elad David, Amit LeVi
Probes on model internals could help monitor agentic systems if they identify harmful text or tool actions before those actions are generated. We ask when an internal readout suppo…
Mirage Probes: How Vision Models Fake Visual Understanding
Daniel Ben-Levi, Judah Goldfeder, Weiliang Zhao +5
Vision-language models (VLMs) can answer image-based questions confidently, and often correctly, even when no image is provided. This mirage behavior inflates benchmark scores with…
Extracting Recurring Vulnerabilities from Black-Box LLM-Generated Software
Tomer Kordonsky, Amit LeVi, Maayan Yamin +2
LLMs are increasingly used for code generation, but their outputs often follow recurring templates that can induce predictable vulnerabilities. We study vulnerability persistence i…
Step-Wise Refusal Dynamics in Autoregressive and Diffusion Language Models
Eliron Rahimi, Elad Hirshel, Rom Himelstein +3
Diffusion language models (DLMs) have recently emerged as a competitive alternative to autoregressive (AR) models, offering parallel decoding, competitive generation quality, and i…
Activation Steering for Masked Diffusion Language Models
Adi Shnaidman, Erin Feiglin, Osher Yaari +3
Masked diffusion language models (MDLMs) generate text via iterative masked-token denoising, enabling mask-parallel decoding and distinct controllability and efficiency tradeoffs f…
You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
Amit LeVi, Raz Lapid, Rom Himelstein +3
Many LLM applications require only narrow capabilities, yet standard post-training quantization (PTQ) methods allocate precision without considering the target task. This can waste…
Silenced Biases: The Dark Side LLMs Learned to Refuse
Rom Himelstein, Amit LeVi, Brit Youngmann +2
Safety-aligned large language models (LLMs) are becoming increasingly widespread, especially in sensitive applications where fairness is essential and biased outputs can cause sign…
Silent Tokens, Loud Effects: Padding in LLMs
Rom Himelstein, Amit LeVi, Yonatan Belinkov +1
Padding tokens are widely used in large language models (LLMs) to equalize sequence lengths during batched inference. While they should be fully masked, implementation errors can c…
Jailbreak Attack Initializations as Extractors of Compliance Directions
Amit Levi, Rom Himelstein, Yaniv Nemcovsky +2
Safety-aligned LLMs respond to prompts with either compliance or refusal, each corresponding to distinct directions in the model's activation space. Recent works show that initiali…