Papers by Andrea Seveso
SFAL: Semantic-Functional Alignment Scores for Distributional Evaluation of Auto-Interpretability in Sparse Autoencoders (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Interpreting the internal representations of large language models (LLMs) is crucial for their deployment in real-world applications, impacting areas such as AI safety, debugging, and compliance. |
| Approach: | They propose an alternative evaluation strategy that assesses the alignment between the semantic neighbourhoods of features and their functional neighbourhoods by using co-occurrence statistics. |
| Outcome: | The proposed evaluation strategy reduces reliance on scoring on large-scale models and improves efficiency and cost-effectiveness. |
Contrastive Explanations of Text Classifiers as a Service (2022.naacl-demo)
Copied to clipboard
| Challenge: | ContrXT provides time contrastive explanations of black box text classifiers by manipulating binary decision diagrams. |
| Approach: | They propose a system that provides time contrastive explanations of black box classifiers as a service by manipulating binary decision diagrams. |
| Outcome: | The proposed system has a throughput of 2.55 users per second and is available as a python pip package. |
SAFE: A Sparse Autoencoder-Based Framework for Robust Query Enrichment and Hallucination Mitigation in LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models suffer from hallucinations, which can undermine their performance in critical applications. |
| Approach: | They propose a framework for detecting and mitigating hallucinations by leveraging SAEs. |
| Outcome: | The proposed framework improves query generation accuracy and mitigates hallucinations across datasets. |
Can Role Vectors Affect LLM Behaviour? (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study has shown that personas influence LLM performance, but their direct impact remains unclear. |
| Approach: | They propose a novel approach to guiding LLM behaviour through role vectors . they construct 29 role vector derived from model activations and evaluate their impact . |
| Outcome: | The proposed approach improves in-domain task performance while yielding unexpected gains. |
ITALIC: An Italian Culture-Aware Natural Language Benchmark (2025.naacl-long)
Copied to clipboard
| Challenge: | ITALIC is a large-scale benchmark dataset of 10,000 multiple-choice questions designed to evaluate the natural language understanding of the Italian language and culture. |
| Approach: | They propose to use a large-scale benchmark dataset to evaluate the natural language understanding of the Italian language and culture. |
| Outcome: | The ITALIC dataset spans 12 domains and uses 17 state-of-the-art LLMs to assess the natural language understanding of the italian language and culture. |