8 papers
Beyond Data Filtering: Knowledge Localization for Capability Removal in LLMs
Igor Shilov, Alex Cloud, Aryo Pradipta Gema +5
Large Language Models increasingly possess capabilities that carry dual-use risks. While data filtering has emerged as a pretraining-time mitigation, it faces significant challenge…
Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them
Neel Rajani, Aryo Pradipta Gema, Seraphina Goldfarb-Tarrant +1
Training large language models (LLMs) for reasoning via maths and code datasets has become a major new focus in LLM post-training. Two particularly popular approaches are reinforce…
Inverse Scaling in Test-Time Compute
Aryo Pradipta Gema, Alexander Hägele, Runjin Chen +11
We construct evaluation tasks where extending the reasoning length of Large Reasoning Models (LRMs) deteriorates performance, exhibiting an inverse scaling relationship between tes…
GRADA: Graph-based Reranking against Adversarial Documents Attack
Jingjie Zheng, Aryo Pradipta Gema, Giwon Hong +4
Retrieval Augmented Generation (RAG) frameworks improve the accuracy of large language models (LLMs) by integrating external knowledge from retrieved documents, thereby overcoming…
Noiser: Bounded Input Perturbations for Attributing Large Language Models
Mohammad Reza Ghasemi Madani, Aryo Pradipta Gema, Gabriele Sarti +3
Feature attribution (FA) methods are common post-hoc approaches that explain how Large Language Models (LLMs) make predictions. Accordingly, generating faithful attributions that r…
An Analysis of Decoding Methods for LLM-based Agents for Faithful Multi-Hop Question Answering
Alexander Murphy, Mohd Sanad Zaki Rizvi, Aden Haussmann +4
Large Language Models (LLMs) frequently produce factually inaccurate outputs - a phenomenon known as hallucination - which limits their accuracy in knowledge-intensive NLP tasks. R…