Alessio Galatolo

Alessio Galatolo, Luca Alberto Rappuoli, Katie Winkle et al.

The recent rise in popularity of large language models (LLMs) has prompted considerable concerns about their moral capabilities. Although considerable effort has been dedicated to aligning LLMs with human moral values, existing benchmarks and evaluations remain largely superficial, typically measuring alignment based on final ethical verdicts rather than explicit moral reasoning. In response, this paper aims to advance the investigation of LLMs' moral capabilities by examining their capacity to function as Artificial Moral Assistants (AMAs), systems envisioned in the philosophical literature to support human moral deliberation. We assert that qualifying as an AMA requires more than what state-of-the-art alignment techniques aim to achieve: not only must AMAs be able to discern ethically problematic situations, they should also be able to actively reason about them, navigating between conflicting values outside of those embedded in the alignment phase. Building on existing philosophical literature, we begin by designing a new formal framework of the specific kind of behaviour an AMA should exhibit, individuating key qualities such as deductive and abductive moral reasoning. Drawing on this theoretical framework, we develop a benchmark to test these qualities and evaluate popular open LLMs against it. Our results reveal considerable variability across models and highlight persistent shortcomings, particularly regarding abductive moral reasoning. Our work connects theoretical philosophy with practical AI evaluation while also emphasising the need for dedicated strategies to explicitly enhance moral reasoning capabilities in LLMs. Code available at https://github.com/alessioGalatolo/AMAeval

Alessio Galatolo, Meriem Beloucif, Katie Winkle

Pretrained Language Models (PLMs) store extensive knowledge within their weights, enabling them to recall vast amount of information. However, relying on this parametric knowledge brings some limitations such as outdated information or gaps in the training data. This work addresses these problems by distinguish between two separate solutions: knowledge editing and knowledge augmentation. We introduce Difference Injection for Efficient Knowledge Augmentation and Editing (DIEKÆ), a new method that decouples knowledge processing from the PLM (LLaMA2-7B, in particular) by adopting a series of encoders. These encoders handle external knowledge and inject it into the PLM layers, significantly reducing computational costs and improving performance of the PLM. We propose a novel training technique for these encoders that does not require back-propagation through the PLM, thus greatly reducing the memory and time required to train them. Our findings demonstrate how our method is faster and more efficient compared to multiple baselines in knowledge augmentation and editing during both training and inference. We have released our code and data at https://github.com/alessioGalatolo/DIEKAE.

Alessio Galatolo, Zhenbang Dai, Katie Winkle et al.

Fine-tuning Large Language Models (LLMs) with first-order methods like back-propagation is computationally intensive. Zeroth-Order (ZO) optimisation uses function evaluations instead of gradients, reducing memory usage, but suffers from slow convergence in high-dimensional models. As a result, ZO research in LLMs has mostly focused on classification, overlooking more complex generative tasks. In this paper, we introduce ZOPrO, a novel ZO algorithm designed for Preference Optimisation in LLMs. We begin by analysing the interplay between policy and reward models during traditional (first-order) Preference Optimisation, uncovering patterns in their relative updates. Guided by these insights, we adapt Simultaneous Perturbation Stochastic Approximation (SPSA) with a targeted sampling strategy to accelerate convergence. Through experiments on summarisation, machine translation, and conversational assistants, we demonstrate that our method consistently enhances reward signals while achieving convergence times comparable to first-order methods. While it falls short of some state-of-the-art methods, our work is the first to apply Zeroth-Order methods to Preference Optimisation in LLMs, going beyond classification tasks and paving the way for a largely unexplored research direction. Code and visualisations are available at https://github.com/alessioGalatolo/VisZOPrO