5 papers
A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models
Hamid Kazemi, Atoosa Chegini, Maria Safi
Safety alignment in language models operates through two mechanistically distinct systems: refusal neurons that gate whether harmful knowledge is expressed, and concept neurons tha…
Reasoning's Razor: Reasoning Improves Accuracy but Can Hurt Recall at Critical Operating Points in Safety and Hallucination Detection
Atoosa Chegini, Hamid Kazemi, Garrett Souza +5
Reasoning has become a central paradigm for large language models (LLMs), consistently boosting accuracy across diverse benchmarks. Yet its suitability for precision-sensitive task…
SALSA: Soup-based Alignment Learning for Stronger Adaptation in RLHF
Atoosa Chegini, Hamid Kazemi, Iman Mirzadeh +5
In Large Language Model (LLM) development, Reinforcement Learning from Human Feedback (RLHF) is crucial for aligning models with human values and preferences. RLHF traditionally re…
Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs
Abhimanyu Hans, Yuxin Wen, Neel Jain +8
Large language models can memorize and repeat their training data, causing privacy and copyright risks. To mitigate memorization, we introduce a subtle modification to the next-tok…
Spotting LLMs With Binoculars: Zero-Shot Detection of Machine-Generated Text
Abhimanyu Hans, Avi Schwarzschild, Valeriia Cherepanova +5
Detecting text generated by modern large language models is thought to be hard, as both LLMs and humans can exhibit a wide range of complex behaviors. However, we find that a score…