Alignment Dynamics in LLM Fine-Tuning

Although Large Language Models (LLMs) achieve strong alignment through supervised fine-tuning and reinforcement learning from human feedback, the alignment is often fragile under subsequent fine-tuning. Existing explanations either attribute alignment fragility to gradient geometry or characterize it as a distributional shift in model outputs, yet few provide a unified account that bridges parameter-space learning dynamics with function-space alignment behavior during fine-tuning. In this work, we introduce a tractable alignment score and derive its closed-form update during fine-tuning, yielding a unified framework for alignment dynamics. Our analysis decomposes alignment updates into two competing components: a \textbf{\color{red!60!black} Rebound Force}, governed jointly by the current alignment state and the narrowness of model distribution, and a \textbf{\color{green!60!black} Driving Force}, determined by how the training distribution aligns with outcome-conditioned posteriors over aligned and non-aligned completions. This decomposition explains why prior alignment can be reversed by later fine-tuning and why narrower posterior structure strengthens such reversal. Moreover, our framework predicts a \textbf{Rehearsal Priming Effect}: prior alignment leaves a latent posterior imprint that amplifies the effective Driving Force upon re-exposure, leading to faster re-alignment. We validate these predictions across safety alignment, emergent misalignment, and sentiment settings, demonstrating consistent alignment reversal and accelerated re-alignment under re-exposure. In addition, controlled experiments in safety alignment confirm the predicted dependence of rebound strength on posterior narrowness. Together, these results provide a unified dynamical perspective on how alignment is disrupted and reactivated during LLM fine-tuning.

Explainable Detection of Depression Status Shifts from User Digital Traces

Every day, users generate digital traces (e.g., social media posts, chats, and online interactions) that are inherently timestamped and may reflect aspects of their mental state. These traces can be organized into temporal trajectories that capture how a user's mental health signals evolve, including phases of improvement, deterioration, or stability. In this work, we propose an explainable framework for detecting and analyzing depression-related status shifts in user digital traces. The approach combines multiple BERT-based models to extract complementary signals across different dimensions (e.g., sentiment, emotion, and depression severity). Such signals are then aggregated over time to construct user-level trajectories that are analyzed to identify meaningful change points. To enhance interpretability, the framework integrates a large language model to generate concise and human-readable reports that describe the evolution of mental-health signals and highlight key transitions. We evaluate the framework on two social media datasets. Results show that the approach produces more coherent and informative summaries than direct LLM-based reporting, achieving higher coverage of user history, stronger temporal coherence, and improved sensitivity to change points. An ablation study confirms the contribution of each component, particularly temporal modeling and segmentation. Overall, the method provides an interpretable view of mental health signals over time, supporting research and decision making without aiming at clinical diagnosis.

Precise Verification of Transformers through ReLU-Catalyzed Abstraction Refinement

Formal verification of transformers has become increasingly important due to their widespread deployment in safety-critical applications. Compared to classic neural networks, the inferences of transformers involve highly complex computations, such as dot products in self-attention layers, rendering their verification extremely difficult. Existing approaches explored over-approximation methods by constructing convex constraints to bound the output ranges of transformers, which can achieve high efficiency. However, they may sacrifice verification precision, and consequently introduce significant approximation error that leads to frequent occurrences of false alarms. In this paper, we propose a transformer verification approach that can achieve improved precision. At the core of our approach is a novel usage of ReLU, by which we represent a precise but non-linear bound for dot products such that we can further exploit the rich body of literature for convex relaxation of ReLU to derive precise bounds. We extend two classic approaches to the context of transformers, a rule-based one and an optimization-based one, resulting in two new frameworks for efficient and precise verification. We evaluate our approaches on different model architectures and robustness properties derived from two datasets about sentiment analysis, and compare with the state-of-the-art baseline approach. Compared to the baseline, our approach can achieve significant precision improvement for most of the verification tasks with acceptable compromise of efficiency, which demonstrates the effectiveness of our approach.

LLMs as annotators of credibility assessment in Danish asylum decisions: evaluating classification performance and errors beyond aggregated metrics

Off-the-shelf large language models (LLMs) are increasingly used to automate text annotation, yet their effectiveness remains underexplored for underrepresented languages and specialized domains where the class definition requires subtle expert understanding. We investigate LLM-based annotation for a novel legal NLP task: identifying the presence and sentiment of credibility assessments in asylum decision texts. We introduce RAB-Cred, a Danish text classification dataset featuring high-quality, expert annotations and valuable metadata such as annotator confidence and asylum case outcome. We benchmark 21 open-weight models and 30 system-user prompt combinations for this task, and systematically evaluate the effect of model and prompt choice for zero-shot and few-shot classification. We zoom in on the errors made by top-performing models and prompts, investigating error consistency across LLMs, inter-class confusion, correlation with human confidence and sample-wise difficulty and severity of LLM mistakes. Our results confirm the potential of LLMs for cost-effective labeling of asylum decisions, but highlight the imperfect and inconsistent nature of LLM annotators, and the need to look beyond the predictions of a single, arbitrarily chosen model. The RAB-Cred dataset and code are available at https://github.com/glhr/RAB-Cred

Continual Learning with Multilingual Foundation Model

This paper presents a multi-stage framework for detecting reclaimed slurs in multilingual social media discourse. It addresses the challenge of identifying reclamatory versus non-reclamatory usage of LGBTQ+-related slurs across English, Spanish, and Italian tweets. The framework handles three intertwined methodological challenges like data scarcity, class imbalance, and cross-linguistic variation in sentiment expression. It integrates data-driven model selection via cross-validation, semantic-preserving augmentation through back-translation, inductive transfer learning with dynamic epoch-level undersampling, and domain-specific knowledge injection via masked language modeling. Eight multilingual embedding models were evaluated systematically, with XLM-RoBERTa selected as the foundation model based on macro-averaged F1 score. Data augmentation via GPT-4o-mini back-translation to alternate languages effectively tripled the training corpus while preserving semantic content and class distribution ratios. The framework produces four final runs for the evaluation purposes where RUN 1 is inductive transfer learning with augmentation and undersampling, RUN 2 with masked language modeling pre-training, RUN 3 and RUN 4 are previous predictions refined via language-specific decision thresholds optimized via ROC analysis. Language-specific threshold refinement reveals that optimal decision boundaries vary significantly across languages. This reflects distributional differences in model confidence scores and linguistic variation in reclamatory language usage. The threshold-based optimization yields 2-5% absolute F1 improvement without requiring model retraining. The methodology is fully reproducible, with all code and experimental setup available at https://github.com/rbg-research/MultiPRIDE-Evalita-2026.

Multimodal Hidden Markov Models for Persistent Emotional State Tracking

Tracking an interpretable emotional arc of a conversation via the sentiment of individual utterances processed as a whole is central to both understanding and guiding communication in applied, especially clinical, conversational contexts. Existing approaches to emotion recognition operate at the utterance level, obscuring the persistent phases that characterize real conversational dynamics. We propose a lightweight framework that models conversational emotion as a sequence of latent emotional regimes using sticky factorial HDP-HMMs over multimodal valence-arousal representations derived from simultaneous video, audio and textual input. We evaluate the quality of regime prediction using LLM-as-a-Judge, geometric, and temporal consistency metrics, demonstrating that the sticky HDP-HMM produces more interpretable regime sequences than the baseline Gaussian HMM at a fraction of the computational cost of LLM-based dialogue state tracking methods. In addition, Question-Answer experiments in a clinical dataset suggest that meaningful emotional phases can reliably be recovered from multimodal valence-arousal trajectories and used to improve the quality of LLM responses in unstable affective regimes via context augmentation. This framework thus opens a path toward interpretable, lightweight, and actionable analysis of conversational emotion dynamics at scale.

Output Composability of QLoRA PEFT Modules for Plug-and-Play Attribute-Controlled Text Generation

Parameter-efficient fine-tuning (PEFT) techniques offer task-specific fine-tuning at a fraction of the cost of full fine-tuning, but require separate fine-tuning for every new task (combination). In this paper, we explore three ways of generalising beyond single-task training/inference: (i) training on combinations of multiple, related datasets; (ii) at inference, composing the weight matrices of separately trained PEFT modules; and (iii) at inference, composing the outputs of separately trained PEFT modules. We test these approaches on three different LLMs, QLoRA as the PEFT technique, and three sets of controlled text generation datasets for sentiment control, topic control, and multi-attribute control. We find that summing PEFT module outputs is a particularly strong composition method, which consistently either outperforms or matches the performance of alternative approaches. This is the case even when comparing against single-task specialised modules on the single-task test set, where three-module output composition achieves an average 2% point performance increase across all models for sentiment control.

Persistent and Conversational Multi-Method Explainability for Trustworthy Financial AI

Financial institutions increasingly require AI explanations that are persistent, cross-validated across methods, and conversationally accessible to human decision-makers. We present an architecture for human-centered explainable AI in financial sentiment analysis that combines three contributions. First, we treat XAI artifacts -- LIME feature attributions, occlusion-based word importance scores, and saliency heatmaps -- as persistent, searchable objects in distributed S3-compatible storage with structured metadata and natural-language summaries, enabling semantic retrieval over explanation history and automatic index reconstruction after system failures. Second, we enable multi-method explanation triangulation, where a retrieval-augmented generation (RAG) assistant compares and synthesizes results from multiple XAI methods applied to the same prediction, allowing users to assess explanation robustness through natural-language dialogue. Third, we evaluate the faithfulness of generated explanations using automated checks over grounding completeness, hallucinated claims, and method-attribution behavior. We demonstrate the architecture on an EXTRA-BRAIN financial sentiment analysis pipeline using FinBERT predictions and present evaluation results showing that constrained prompting reduces hallucination rate by 36\% and increases method-attribution citations by 73\% compared to naive prompting. We discuss implications for trustworthy, human-centered AI services in regulated financial environments.

Correcting Selection Bias in Sparse User Feedback for Large Language Model Quality Estimation: A Multi-Agent Hierarchical Bayesian Approach

[Abridged] Production LLM deployments receive feedback from a non-random fraction of users: thumbs sit mostly in the tails of the satisfaction distribution, and a naive average over them can land 40-50 percentage points away from true system quality. We treat this as a topic- and sentiment- stratified selection-bias problem and propose a three-agent hierarchical Bayesian pipeline that does not require ground-truth labels on individual interactions. A Topic Clustering Agent partitions the stream via UMAP + HDBSCAN over text embeddings; a Bias Modeling Agent fits a two-stage hierarchical Beta-Binomial under NUTS, inferring per-topic selection rates $s_c$ and quality $q_c$ with partial pooling; a Synthesis Agent reweights $q_c$ by true topic prevalence $\hatπ_c = n_c/N$ to report a bias-corrected aggregate posterior $\bar Q = \sum_c \hatπ_c q_c$ with credible interval, plus drift signals for online recalibration. Validation uses UltraFeedback (N=10,232 retained interactions, $C=18$ clusters, $Q^\star=0.6249$) with simulated topic- and sentiment-dependent selection biases. We compare five Bayesian variants against Naive and IPW baselines. A mild prior on the feedback channel (typical positive-feedback rate and negative-to-positive ratio, both readable from any production dashboard without labels) keeps Hierarchical-Informed within 4-13 pp of $Q^\star$ as the bias ratio sweeps from 1:1 to 30:1, with 95% credible intervals covering $Q^\star$ in 50/50 random-seed replicates at $κ_{\max}=10$. Without channel-side priors, every weak-prior variant misses $Q^\star$ by 22-33 pp: the per-cluster sufficient statistics admit a one-parameter family of equally good fits, and the prior on the bias channel (not on latent quality) is what breaks the degeneracy.

The Algorithmic Caricature: Auditing LLM-Generated Political Discourse Across Crisis Events

Large Language Models (LLMs) can generate fluent political text at scale, raising concerns about synthetic discourse during crises and social conflict. Existing AI-text detection often focuses on sentence-level cues such as perplexity, burstiness, or token irregularities, but these signals may weaken as generative systems improve. We instead adopt a Computational Social Science perspective and ask whether synthetic political discourse behaves like an observed online population. We construct a paired corpus of 1,789,406 posts across nine crisis events: COVID-19, the Jan. 6 Capitol attack, the 2020 and 2024 U.S. elections, Dobbs/Roe v. Wade, the 2020 BLM protests, U.S. midterms, the Utah shooting, and the U.S.-Iran war. For each event, we compare observed discourse from social platforms with synthetic discourse generated for the same context. We evaluate four dimensions: emotional intensity, structural regularity, lexical-ideological framing, and cross-event dependency, using mean gaps and dispersion evidence. Across events, synthetic discourse is fluent but population-level unrealistic. It is generally more negative and less dispersed in sentiment, structurally more regular, and lexically more abstract than observed discourse. Observed discourse instead shows broader emotional variation, longer-tailed structural distributions, and more context-specific, colloquial lexical markers. These differences are event-dependent: larger for fast-moving, decentralized crises and smaller for formal or institutionally mediated events. We summarize them with a simple event-level measure, the Caricature Gap. Our findings suggest that the main limitation of synthetic political discourse is not grammar or fluency, but reduced population realism. Population-level auditing complements traditional text-detection and provides a CSS framework for evaluating the social realism of generated discourse.