Philip Lippmann

Andrea Tocchetti, Lorenzo Corti, Agathe Balayn et al.

Despite the impressive performance of Artificial Intelligence (AI) systems, their robustness remains elusive and constitutes a key issue that impedes large-scale adoption. Robustness has been studied in many domains of AI, yet with different interpretations across domains and contexts. In this work, we systematically survey the recent progress to provide a reconciled terminology of concepts around AI robustness. We introduce three taxonomies to organize and describe the literature both from a fundamental and applied point of view: 1) robustness by methods and approaches in different phases of the machine learning pipeline; 2) robustness for specific model architectures, tasks, and systems; and in addition, 3) robustness assessment methodologies and insights, particularly the trade-offs with other trustworthiness properties. Finally, we identify and discuss research gaps and opportunities and give an outlook on the field. We highlight the central role of humans in evaluating and enhancing AI robustness, considering the necessary knowledge humans can provide, and discuss the need for better understanding practices and developing supportive tools in the future.

Philip Lippmann, Matthijs T. J. Spaan, Jie Yang

Intelligent agents designed for interactive environments face significant challenges in text-based games, a domain that demands complex reasoning and adaptability. While agents based on large language models (LLMs) using self-reflection have shown promise, they struggle when initially successful and exhibit reduced effectiveness when using smaller LLMs. We introduce Sweet&Sour, a novel approach that addresses these limitations in existing reflection methods by incorporating positive experiences and managed memory to enrich the context available to the agent at decision time. Our comprehensive analysis spans both closed- and open-source LLMs and demonstrates the effectiveness of Sweet&Sour in improving agent performance, particularly in scenarios where previous approaches fall short.

Philip Lippmann, Konrad Skublicki, Joshua Tanner et al.

Due to the significant time and effort required for handcrafting translations, most manga never leave the domestic Japanese market. Automatic manga translation is a promising potential solution. However, it is a budding and underdeveloped field and presents complexities even greater than those found in standard translation due to the need to effectively incorporate visual elements into the translation process to resolve ambiguities. In this work, we investigate to what extent multimodal large language models (LLMs) can provide effective manga translation, thereby assisting manga authors and publishers in reaching wider audiences. Specifically, we propose a methodology that leverages the vision component of multimodal LLMs to improve translation quality and evaluate the impact of translation unit size, context length, and propose a token efficient approach for manga translation. Moreover, we introduce a new evaluation dataset -- the first parallel Japanese-Polish manga translation dataset -- as part of a benchmark to be used in future research. Finally, we contribute an open-source software suite, enabling others to benchmark LLMs for manga translation. Our findings demonstrate that our proposed methods achieve state-of-the-art results for Japanese-English translation and set a new standard for Japanese-Polish.

Aleksander Buszydlik, Karol Dobiczek, Michał Teodor Okoń et al.

We consider the problem of red teaming LLMs on elementary calculations and algebraic tasks to evaluate how various prompting techniques affect the quality of outputs. We present a framework to procedurally generate numerical questions and puzzles, and compare the results with and without the application of several red teaming techniques. Our findings suggest that even though structured reasoning and providing worked-out examples slow down the deterioration of the quality of answers, the gpt-3.5-turbo and gpt-4 models are not well suited for elementary calculations and reasoning tasks, also when being red teamed.

Philip Lippmann, Jie Yang

Specialized reasoning language models (RLMs) have demonstrated that scaling test-time computation through detailed reasoning traces significantly enhances performance. Although these traces effectively facilitate knowledge distillation into smaller, instruction-tuned models, the precise nature of transferred reasoning remains unclear. In this study, we investigate to what extent distilled models internalize replicated stylistic patterns during reasoning. To this end, we systematically analyze reasoning traces, identifying structural and lexical patterns that characterize successful reasoning. We then introduce two new datasets -- a dataset of emergent reasoning traces and a synthetic dataset explicitly constructed to replicate these stylistic patterns -- to precisely examine their influence on distilled models' reasoning capabilities. We find that models trained on the synthetic traces achieve comparable performance, indicating that distilled reasoning abilities rely significantly on surface-level patterns. Surprisingly, we observe an increase in performance even when the synthetic traces are altered to lead to the wrong answer. Our findings highlight how stylistic patterns can be leveraged to efficiently enhance LM reasoning across diverse model families.

Philip Lippmann, Matthijs T. J. Spaan, Jie Yang

Language models (LMs) have achieved impressive accuracy across a variety of tasks but remain vulnerable to high-confidence misclassifications, also referred to as unknown unknowns (UUs). These UUs cluster into blind spots in the feature space, leading to significant risks in high-stakes applications. This is particularly relevant for smaller, lightweight LMs that are more susceptible to such errors. While the identification of UUs has been extensively studied, their mitigation remains an open challenge, including how to use identified UUs to eliminate unseen blind spots. In this work, we propose a novel approach to address blind spot mitigation through the use of intelligent agents -- either humans or large LMs -- as teachers to characterize UU-type errors. By leveraging the generalization capabilities of intelligent agents, we identify patterns in high-confidence misclassifications and use them to generate targeted synthetic samples to improve model robustness and reduce blind spots. We conduct an extensive evaluation of our method on three classification tasks and demonstrate its effectiveness in reducing the number of UUs, all while maintaining a similar level of accuracy. We find that the effectiveness of human computation has a high ceiling but is highly dependent on familiarity with the underlying task. Moreover, the cost gap between humans and LMs surpasses an order of magnitude, as LMs attain human-like generalization and generation performance while being more scalable.

Philip Lippmann, Jie Yang

Context-aware embedding methods boost retrieval accuracy by conditioning on corpus statistics (e.g., term co-occurrence and topical patterns) extracted from neighboring documents. However, this context-aware approach requires access to the target corpus or requires domain-specific finetuning, posing practical barriers in privacy-sensitive or resource-constrained settings. We present ZEST, a zero-shot contextual adaptation framework that replaces real corpus access with a one-time offline synthesis of a compact proxy. Given only a handful exemplar documents representative of the general target domain, we use a multi-step hierarchical procedure to generate a synthetic context corpus of several hundred documents that aims to emulate key domain-specific distributions. At inference, the frozen context-aware encoder uses this proxy corpus -- without any finetuning or target corpus access -- to produce domain-adapted embeddings. Across the MTEB benchmark, ZEST's zero-shot synthetic context adaptation using only five example documents performs within 0.5% of models leveraging full target corpus access -- demonstrating remarkable efficacy without any retraining. ZEST thus provides a practical method for deploying high-performance, adaptable embeddings in constrained environments.