Lisa Alazraki

Lisa Alazraki, Ali Ghachem, Neophytos Polydorou et al.

In this work, we present a new dataset and a computational strategy for a digital coach that aims to guide users in practicing the protocols of self-attachment therapy. Our framework augments a rule-based conversational agent with a deep-learning classifier for identifying the underlying emotion in a user's text response, as well as a deep-learning assisted retrieval method for producing novel, fluent and empathetic utterances. We also craft a set of human-like personas that users can choose to interact with. Our goal is to achieve a high level of engagement during virtual therapy sessions. We evaluate the effectiveness of our framework in a non-clinical trial with N=16 participants, all of whom have had at least four interactions with the agent over the course of five days. We find that our platform is consistently rated higher for empathy, user engagement and usefulness than the simple rule-based framework. Finally, we provide guidelines to further improve the design and performance of the application, in accordance with the feedback received.

Lisa Alazraki, Lluis Castrejon, Mostafa Dehghani et al.

This paper studies ensembling in the era of Large Vision-Language Models (LVLMs). Ensembling is a classical method to combine different models to get increased performance. In the recent work on Encyclopedic-VQA the authors examine a wide variety of models to solve their task: from vanilla LVLMs, to models including the caption as extra context, to models augmented with Lens-based retrieval of Wikipedia pages. Intuitively these models are highly complementary, which should make them ideal for ensembling. Indeed, an oracle experiment shows potential gains from 48.8% accuracy (the best single model) all the way up to 67% (best possible ensemble). So it is a trivial exercise to create an ensemble with substantial real gains. Or is it?

Alicia Jiayun Law, Ruoyu Hu, Lisa Alazraki et al.

In this work, we propose a computational framework that leverages existing out-of-language data to create a conversational agent for the delivery of Self-Attachment Technique (SAT) in Mandarin. Our framework does not require large-scale human translations, yet it achieves a comparable performance whilst also maintaining safety and reliability. We propose two different methods of augmenting available response data through empathetic rewriting. We evaluate our chatbot against a previous, English-only SAT chatbot through non-clinical human trials (N=42), each lasting five days, and quantitatively show that we are able to attain a comparable level of performance to the English SAT chatbot. We provide qualitative analysis on the limitations of our study and suggestions with the aim of guiding future improvements.

Sina Elahimanesh, Shayan Salehi, Sara Zahedi Movahed et al.

In the wake of the post-pandemic era, marked by social isolation and surging rates of depression and anxiety, conversational agents based on digital psychotherapy can play an influential role compared to traditional therapy sessions. In this work, we develop a voice-capable chatbot in Farsi to guide users through Self-Attachment (SAT), a novel, self-administered, holistic psychological technique based on attachment theory. Our chatbot uses a dynamic array of rule-based and classification-based modules to comprehend user input throughout the conversation and navigates a dialogue flowchart accordingly, recommending appropriate SAT exercises that depend on the user's emotional and mental state. In particular, we collect a dataset of over 6,000 utterances and develop a novel sentiment-analysis module that classifies user sentiment into 12 classes, with accuracy above 92%. To keep the conversation novel and engaging, the chatbot's responses are retrieved from a large dataset of utterances created with the aid of Farsi GPT-2 and a reinforcement learning approach, thus requiring minimal human annotation. Our chatbot also offers a question-answering module, called SAT Teacher, to answer users' questions about the principles of Self-Attachment. Finally, we design a cross-platform application as the bot's user interface. We evaluate our platform in a ten-day human study with N=52 volunteers from the non-clinical population, who have had over 2,000 dialogues in total with the chatbot. The results indicate that the platform was engaging to most users (75%), 72% felt better after the interactions, and 74% were satisfied with the SAT Teacher's performance.

Lisa Alazraki, Marek Rei

External tools help large language models succeed at tasks where they would otherwise typically fail. In existing frameworks, choosing tools at test time relies on naive greedy decoding, regardless of whether the model has been fine-tuned on tool-annotated data or prompted with in-context examples. In contrast, we find that gathering and choosing among a suitable set of candidate tools has greater potential to lead to an optimal selection. We present Tool selECTion via meta-reasONing (TECTON), a two-phase system that first reasons over a task and outputs candidate tools using a custom fine-tuned language modelling head. Then, with the custom head disabled, it meta-reasons (i.e., it reasons over the previous reasoning process) to make a final choice. We show that TECTON results in substantial gains--both in-distribution and out-of-distribution--on a range of math reasoning datasets.

Aryan Agrawal, Lisa Alazraki, Shahin Honarvar et al.

Large Language Models (LLMs) are highly vulnerable to input perturbations, as even a small prompt change may result in a substantially different output. Existing methods to enhance LLM robustness are primarily focused on perturbed data samples, whereas improving resiliency to perturbations of task-level instructions has remained relatively underexplored. In this work, we focus on character- and word-level edits of task-specific instructions, which substantially degrade downstream performance. We experiment with a variety of techniques to enhance the robustness of LLMs, including self-denoising and representation alignment, testing different models (Llama 3 and Flan-T5), datasets (CoLA, QNLI, SST-2) and instructions (both task-oriented and role-oriented). We find that, on average, self-denoising -- whether performed by a frozen LLM or a fine-tuned model -- achieves substantially higher performance gains than alternative strategies, including more complex baselines such as ensembling and supervised methods.

Lisa Alazraki, Maximilian Mozes, Jon Ander Campos et al.

Showing incorrect answers to Large Language Models (LLMs) is a popular strategy to improve their performance in reasoning-intensive tasks. It is widely assumed that, in order to be helpful, the incorrect answers must be accompanied by comprehensive rationales, explicitly detailing where the mistakes are and how to correct them. However, in this work we present a counterintuitive finding: we observe that LLMs perform better in math reasoning tasks when these rationales are eliminated from the context and models are left to infer on their own what makes an incorrect answer flawed. This approach also substantially outperforms chain-of-thought prompting in our evaluations. These results are consistent across LLMs of different sizes and varying reasoning abilities. To gain an understanding of why LLMs learn from mistakes more effectively without explicit corrective rationales, we perform a thorough analysis, investigating changes in context length and answer diversity between different prompting strategies, and their effect on performance. We also examine evidence of overfitting to the in-context rationales when these are provided, and study the extent to which LLMs are able to autonomously infer high-quality corrective rationales given only incorrect answers as input. We find evidence that, while incorrect answers are more beneficial for LLM learning than additional diverse correct answers, explicit corrective rationales over-constrain the model, thus limiting those benefits.

William F. Shen, Xinchi Qiu, Chenxi Whitehouse et al.

Recently, rubrics have been used to guide LLM judges in capturing subjective, nuanced, multi-dimensional human preferences, and have been extended from evaluation to reward signals for reinforcement fine-tuning (RFT). However, rubric generation remains hard to control: rubrics often lack coverage, conflate dimensions, misalign preference direction, and contain redundant or highly correlated criteria, degrading judge accuracy and producing suboptimal rewards during RFT. We propose RRD, a principled framework for rubric refinement built on a recursive decompose-filter cycle. RRD decomposes coarse rubrics into fine-grained, discriminative criteria, expanding coverage while sharpening separation between responses. A complementary filtering mechanism removes misaligned and redundant rubrics, and a correlation-aware weighting scheme prevents over-representing highly correlated criteria, yielding rubric sets that are informative, comprehensive, and non-redundant. Empirically, RRD delivers large, consistent gains across both evaluation and training: it improves preference-judgment accuracy on JudgeBench and PPE for both GPT-4o and Llama3.1-405B judges, achieving top performance in all settings with up to +17.7 points on JudgeBench. When used as the reward source for RFT on WildChat, it yields substantially stronger and more stable learning signals, boosting reward by up to 160% (Qwen3-4B) and 60% (Llama3.1-8B) versus 10-20% for prior rubric baselines, with gains that transfer to HealthBench-Hard and BiGGen Bench. Overall, RRD establishes recursive rubric refinement as a scalable and interpretable foundation for LLM judging and reward modeling in open-ended domains.

Lisa Alazraki, William F. Shen, Yoram Bachrach et al.

Small language models are increasingly viewed as a promising, cost-effective approach to agentic AI, with proponents claiming they are sufficiently capable for agentic workflows. However, while smaller agents can closely match larger ones on simple tasks, it remains unclear how their performance scales with task complexity, when large models become necessary, and how to better leverage small agents for long-horizon workloads. In this work, we empirically show that small agents' performance fails to scale with task complexity on deep search and coding tasks, and we introduce Strategy Auctions for Workload Efficiency (SALE), an agent framework inspired by freelancer marketplaces. In SALE, agents bid with short strategic plans, which are scored by a systematic cost-value mechanism and refined via a shared auction memory, enabling per-task routing and continual self-improvement without training a separate router or running all models to completion. Across deep search and coding tasks of varying complexity, SALE reduces reliance on the largest agent by 53%, lowers overall cost by 35%, and consistently improves upon the largest agent's pass@1 with only a negligible overhead beyond executing the final trace. In contrast, established routers that rely on task descriptions either underperform the largest agent or fail to reduce cost -- often both -- underscoring their poor fit for agentic workflows. These results suggest that while small agents may be insufficient for complex workloads, they can be effectively "scaled up" through coordinated task allocation and test-time self-improvement. More broadly, they motivate a systems-level view of agentic AI in which performance gains come less from ever-larger individual models and more from market-inspired coordination mechanisms that organize heterogeneous agents into efficient, adaptive ecosystems.

Tyler A. Chang, Catherine Arnett, Abdelrahman Eldesokey et al. · uw

To date, there exist almost no culturally-specific evaluation benchmarks for large language models (LLMs) that cover a large number of languages and cultures. In this paper, we present Global PIQA, a participatory commonsense reasoning benchmark for over 100 languages, constructed by hand by 335 researchers from 65 countries around the world. The 116 language varieties in Global PIQA cover five continents, 14 language families, and 23 writing systems. In the non-parallel split of Global PIQA, over 50% of examples reference local foods, customs, traditions, or other culturally-specific elements. We find that state-of-the-art LLMs perform well on Global PIQA in aggregate, but they exhibit weaker performance in lower-resource languages (up to a 37% accuracy gap, despite random chance at 50%). Open models generally perform worse than proprietary models. Global PIQA highlights that in many languages and cultures, everyday knowledge remains an area for improvement, alongside more widely-discussed capabilities such as complex reasoning and expert knowledge. Beyond its uses for LLM evaluation, we hope that Global PIQA provides a glimpse into the wide diversity of cultures in which human language is embedded.

Marco De Santis, Lisa Alazraki

This paper presents our submission to the MRL 2025 Shared Task on Multilingual Physical Reasoning Datasets. The objective of the shared task is to create manually-annotated evaluation data in the physical commonsense reasoning domain, for languages other than English, following a format similar to PIQA. Our contribution, FormaMentis, is a novel benchmark for physical commonsense reasoning that is grounded in Italian language and culture. The data samples in FormaMentis are created by expert annotators who are native Italian speakers and are familiar with local customs and norms. The samples are additionally translated into English, while preserving the cultural elements unique to the Italian context.

Lisa Alazraki, Lihu Chen, Ana Brassard et al.

Large Language Models (LLMs) have achieved high accuracy on complex commonsense and mathematical problems that involve the composition of multiple reasoning steps. However, current compositional benchmarks testing these skills tend to focus on either commonsense or math reasoning, whereas LLM agents solving real-world tasks would require a combination of both. In this work, we introduce an Agentic Commonsense and Math benchmark (AgentCoMa), where each compositional task requires a commonsense reasoning step and a math reasoning step. We test it on 61 LLMs of different sizes, model families, and training strategies. We find that LLMs can usually solve both steps in isolation, yet their accuracy drops by ~30% on average when the two are combined. This is a substantially greater performance gap than the one we observe in prior compositional benchmarks that combine multiple steps of the same reasoning type. In contrast, non-expert human annotators can solve the compositional questions and the individual steps in AgentCoMa with similarly high accuracy. Furthermore, we conduct a series of interpretability studies to better understand the performance gap, examining neuron patterns, attention maps and membership inference. Our work underscores a substantial degree of model brittleness in the context of mixed-type compositional reasoning and offers a test bed for future improvement.

Lisa Alazraki, Tan Yi-Chern, Jon Ander Campos et al.

The capabilities of Large Language Models (LLMs) are routinely evaluated by other LLMs trained to predict human preferences. This framework--known as LLM-as-a-judge--is highly scalable and relatively low cost. However, it is also vulnerable to malicious exploitation, as LLM responses can be tuned to overfit the preferences of the judge. Previous work shows that the answers generated by a candidate-LLM can be edited post hoc to maximise the score assigned to them by a judge-LLM. In this study, we adopt a different approach and use the signal provided by judge-LLMs as a reward to adversarially tune models that generate text preambles designed to boost downstream performance. We find that frozen LLMs pipelined with these models attain higher LLM-evaluation scores than existing frameworks. Crucially, unlike other frameworks which intervene directly on the model's response, our method is virtually undetectable. We also demonstrate that the effectiveness of the tuned preamble generator transfers when the candidate-LLM and the judge-LLM are replaced with models that are not used during training. These findings raise important questions about the design of more reliable LLM-as-a-judge evaluation settings. They also demonstrate that human preferences can be reverse engineered effectively, by pipelining LLMs to optimise upstream preambles via reinforcement learning--an approach that could find future applications in diverse tasks and domains beyond adversarial attacks.

Mingxue Xu, Lisa Alazraki, Danilo P. Mandic

Pruning assumes a subnetwork exists in the original deep neural network, which can achieve comparative model performance with less computation than the original. However, it is unclear how the model performance varies with the different subnetwork extractions. In this paper, we choose the representation dimension (or embedding dimension, model dimension, the dimension of the residual stream in the relevant literature) as the entry point to this issue. We investigate the linear transformations in the LLM transformer blocks and consider a specific structured pruning approach, SliceGPT, to extract the subnetworks of different representation dimensions. We mechanistically analyse the activation flow during the model forward passes, and find the representation dimension dominates the linear transformations, model predictions, and, finally, the model performance. Explicit analytical relations are given to calculate the pruned model performance (perplexity and accuracy) without actual evaluation, and are empirically validated with Llama-3-8B-Instruct and Phi-3-mini-4k-Instruct.