Josef Pichlmeier

Dasol Choi, Eugenia Kim, Jaewon Noh et al.

Current LLM safety benchmarks are predominantly English-centric and often rely on translation, failing to capture country-specific harms. Moreover, they rarely evaluate a model's ability to detect culturally embedded sensitivities as distinct from universal harms. We introduce XL-SafetyBench. a suite of 5,500 test cases across 10 country-language pairs, comprising a Jailbreak Benchmark of country-grounded adversarial prompts and a Cultural Benchmark where local sensitivities are embedded within innocuous requests. Each item is constructed via a multi-stage pipeline that combines LLM-assisted discovery, automated validation gates, and dual independent native-speaker annotators per country. To distinguish principled refusal from comprehension failure, we evaluate Attack Success Rate (ASR) alongside two complementary metrics we introduce: Neutral-Safe Rate (NSR) and Cultural Sensitivity Rate (CSR). Evaluating 10 frontier and 27 local LLMs reveals two key findings. First, jailbreak robustness and cultural awareness do not show a coupled relationship among frontier models, so a composite safety score obscures per-axis variation. Second, local models exhibit a near-linear ASR-NSR trade-off (r = -0.81), indicating that their apparent safety reflects generation failure rather than genuine alignment. XL-SafetyBench enables more nuanced, cross-cultural safety evaluation in the multilingual era.

Jakob Sturm, Josef Pichlmeier, Christian Bernhard et al.

Large Language Models (LLMs) are increasingly employed in enterprise question-answering (QA) systems, requiring adaptation to domain-specific knowledge. Among the most prevalent methods for incorporating such knowledge are Retrieval-Augmented Generation (RAG) and fine-tuning (FT). Yet, from a cost-accuracy trade-off perspective, it remains unclear which approach best suits industry scenarios. This study examines the impact of RAG and FT on two closed datasets specific to the automotive industry, assessing answer quality and operational costs. We extend the Cost-of-Pass framework proposed by Erol et al. (arXiv:2504.13359) to jointly assess output quality, generation cost, and user interaction cost. Our findings reveal that while premium models perform best out of the box, open-source models can achieve comparable quality when enhanced with RAG. Overall, RAG emerges as the most effective and cost-efficient adaptation method for both closed- and open-source models.

Ryan Marinelli, Josef Pichlmeier, Tamas Bisztray

In this work, we propose a metric called Number of Thoughts (NofT) to determine the difficulty of tasks pre-prompting and support Large Language Models (LLMs) in production contexts. By setting thresholds based on the number of thoughts, this metric can discern the difficulty of prompts and support more effective prompt routing. A 2% decrease in latency is achieved when routing prompts from the MathInstruct dataset through quantized, distilled versions of Deepseek with 1.7 billion, 7 billion, and 14 billion parameters. Moreover, this metric can be used to detect adversarial prompts used in prompt injection attacks with high efficacy. The Number of Thoughts can inform a classifier that achieves 95% accuracy in adversarial prompt detection. Our experiments ad datasets used are available on our GitHub page: https://github.com/rymarinelli/Number_Of_Thoughts/tree/main.

Josef Pichlmeier, Philipp Ross, Andre Luckow

Large Language Models (LLMs) have experienced widespread adoption across scientific and industrial domains due to their versatility and utility for diverse tasks. Nevertheless, deploying and serving these models at scale with optimal throughput and latency remains a significant challenge, primarily because of LLMs' high computational and memory demands. Specialized models optimized for specific tasks can be combined through a routing mechanism to address these challenges, creating a modular inference system. This paper introduces Expert Router, a scalable routing architecture that directs prompts to specialized expert models. We characterize multiple Expert Router configurations, including different LLama 3 models with quantized and non-quantized weights under up to 1,000 concurrent users. Our findings reveal that Expert Router introduces minimal latency overhead, with the configuration of expert models being a dominating factor in performance outcomes. High-parameter expert models deliver stable throughput and latency under moderate concurrency levels. In contrast, smaller expert models maintain competitive performance across a wider range of concurrent users compared to tensor-parallelized baseline models. This highlights the potential of Expert Router for efficient and scalable LLM deployment.

Thomas Hubregtsen, Josef Pichlmeier, Patrick Stecher et al.

An active area of investigation in the search for quantum advantage is Quantum Machine Learning. Quantum Machine Learning, and Parameterized Quantum Circuits in a hybrid quantum-classical setup in particular, could bring advancements in accuracy by utilizing the high dimensionality of the Hilbert space as feature space. But is the ability of a quantum circuit to uniformly address the Hilbert space a good indicator of classification accuracy? In our work, we use methods and quantifications from prior art to perform a numerical study in order to evaluate the level of correlation. We find a strong correlation between the ability of the circuit to uniformly address the Hilbert space and the achieved classification accuracy for circuits that entail a single embedding layer followed by 1 or 2 circuit designs. This is based on our study encompassing 19 circuits in both 1 and 2 layer configuration, evaluated on 9 datasets of increasing difficulty. Future work will evaluate if this holds for different circuit designs.

Thomas Hubregtsen, Christoph Segler, Josef Pichlmeier et al.

Quantum computers hold great promise for accelerating computationally challenging algorithms on noisy intermediate-scale quantum (NISQ) devices in the upcoming years. Much attention of the current research is directed to algorithmic research on artificial data that is disconnected from live systems, such as optimization of systems or training of learning algorithms. In this paper we investigate the integration of quantum systems into industry-grade system architectures. In this work we propose a system architecture for the integration of quantum accelerators. In order to evaluate our proposed system architecture we implemented various algorithms including a classical system, a gate-based quantum accelerator and a quantum annealer. This algorithm automates user habits using data-driven functions trained on real-world data. This also includes an evaluation of the quantum enhanced kernel, that previously was only evaluated on artificial data. In our evaluation, we showed that the quantum-enhanced kernel performs at least equally well to a classical state-of-the-art kernel. We also showed a low reduction in accuracy and latency numbers within acceptable bounds when running on the gate-based IBM quantum accelerator. We, therefore, conclude it is feasible to integrate NISQ-era devices in industry-grade system architecture in preparation for future hardware improvements.