Lars Klein

Martin Josifoski, Lars Klein, Maxime Peyrard et al.

Recent advances in artificial intelligence (AI) have produced highly capable and controllable systems. This creates unprecedented opportunities for structured reasoning as well as collaboration among multiple AI systems and humans. To fully realize this potential, it is essential to develop a principled way of designing and studying such structured interactions. For this purpose, we introduce the conceptual framework Flows. Flows are self-contained building blocks of computation, with an isolated state, communicating through a standardized message-based interface. This modular design simplifies the process of creating Flows by allowing them to be recursively composed into arbitrarily nested interactions and is inherently concurrency-friendly. Crucially, any interaction can be implemented using this framework, including prior work on AI-AI and human-AI interactions, prompt engineering schemes, and tool augmentation. We demonstrate the potential of Flows on competitive coding, a challenging task on which even GPT-4 struggles. Our results suggest that structured reasoning and collaboration substantially improve generalization, with AI-only Flows adding +21 and human-AI Flows adding +54 absolute points in terms of solve rate. To support rapid and rigorous research, we introduce the aiFlows library embodying Flows. The aiFlows library is available at https://github.com/epfl-dlab/aiflows. Data and Flows for reproducing our experiments are available at https://github.com/epfl-dlab/cc_flows.

Bardia Mohammadi, Lars Klein, Akhil Arora et al.

Tool-augmented language agents speculatively issue likely future tool calls to hide latency, but those calls leak inferred user intent to external services before the agent commits to the branch. Every external observer that received the call retains the disclosure after the agent abandons the branch. Timing is the issue, not authorization: no commit-time cleanup, read-only restriction, or access-control allow-list unsends what an observer already holds. We call these invocations ghost tool calls and propose Speculative Tool Privacy Contracts, a runtime abstraction that treats observation before commitment as a first-class effect, distinct from state mutation. We implement the contracts in a prototype runtime and evaluate twelve policies across three corpora. Speculative dispatch increases what an observer can infer about user intent; post-hoc filters, read-only restrictions, and access-control allow-lists leave that inference intact; only issue-time policies that change or suppress the speculative call's argument or destination projection before dispatch reduce it.

Yusuf Kesmen, Fay Elhassan, Jiayi Ma et al.

Large language models are increasingly deployed as autonomous diagnostic agents, yet they conflate two fundamentally different capabilities: natural-language communication and probabilistic reasoning. We argue that this conflation is an architectural flaw, not an engineering shortcoming. We introduce BMBE (Bayesian Medical Belief Engine), a modular diagnostic dialogue framework that enforces a strict separation between language and reasoning: an LLM serves only as a sensor, parsing patient utterances into structured evidence and verbalising questions, while all diagnostic inference resides in a deterministic, auditable Bayesian engine. Because patient data never enters the LLM, the architecture is private by construction; because the statistical backend is a standalone module, it can be replaced per target population without retraining. This separation yields three properties no autonomous LLM can offer: calibrated selective diagnosis with a continuously adjustable accuracy-coverage tradeoff, a statistical separation gap where even a cheap sensor paired with the engine outperforms a frontier standalone model from the same family at a fraction of the cost, and robustness to adversarial patient communication styles that cause standalone doctors to collapse. We validate across empirical and LLM-generated knowledge bases against frontier LLMs, confirming the advantage is architectural, not informational.

Daniyar Chumbalov, Lars Klein, Lucas Maystre et al.

A comparison-based search algorithm lets a user find a target item $t$ in a database by answering queries of the form, ``Which of items $i$ and $j$ is closer to $t$?'' Instead of formulating an explicit query (such as one or several keywords), the user navigates towards the target via a sequence of such (typically noisy) queries. We propose a scale-free probabilistic oracle model called $γ$-CKL for such similarity triplets $(i,j;t)$, which generalizes the CKL triplet model proposed in the literature. The generalization affords independent control over the discriminating power of the oracle and the dimension of the feature space containing the items. We develop a search algorithm with provably exponential rate of convergence under the $γ$-CKL oracle, thanks to a backtracking strategy that deals with the unavoidable errors in updating the belief region around the target. We evaluate the performance of the algorithm both over the posited oracle and over several real-world triplet datasets. We also report on a comprehensive user study, where human subjects navigate a database of face portraits.

Nearchos Potamitis, Lars Klein, Akhil Arora

Large language models (LLMs) are increasingly deployed in settings where reasoning, such as multi-step problem solving and chain-of-thought, is essential. Yet, current evaluation practices overwhelmingly report single-run accuracy while ignoring the intrinsic uncertainty that naturally arises from stochastic decoding. This omission creates a blind spot because practitioners cannot reliably assess whether a method's reported performance is stable, reproducible, or cost-consistent. We introduce ReasonBENCH, the first benchmark designed to quantify the underlying instability in LLM reasoning. ReasonBENCH provides (i) a modular evaluation library that standardizes reasoning frameworks, models, and tasks, (ii) a multi-run protocol that reports statistically reliable metrics for both quality and cost, and (iii) a public leaderboard to encourage variance-aware reporting. Across tasks from different domains, we find that the vast majority of reasoning strategies and models exhibit high instability. Notably, even strategies with similar average performance can display confidence intervals up to four times wider, and the top-performing methods often incur higher and less stable costs. Such instability compromises reproducibility across runs and, consequently, the reliability of reported performance. To better understand these dynamics, we further analyze the impact of prompts, model families, and scale on the trade-off between solve rate and stability. Our results highlight reproducibility as a critical dimension for reliable LLM reasoning and provide a foundation for future reasoning methods and uncertainty quantification techniques. ReasonBENCH is publicly available at https://github.com/au-clan/ReasonBench .

Xavier Theimer-Lienhard, Mushtaha El-Amin, Fay Elhassan et al.

Clinical decision support systems (CDSS) require scrutable, auditable pipelines that enable rigorous, reproducible validation. Yet current LLM-based CDSS remain largely opaque. Most "open" models are open-weight only, releasing parameters while withholding the data provenance, curation procedures, and generation pipelines that determine model behavior. Fully Open (FO) models, which expose the complete training stack end-to-end, do not currently exist in medicine. We introduce Fully Open Meditron, the first fully open pipeline for building LLM-CDSS, comprising a clinician-audited training corpus, a reproducible data construction and training framework, and a use-aligned evaluation protocol. The corpus unifies eight public medical QA datasets into a normalized conversational format and expands coverage with three clinician-vetted synthetic extensions: exam-style QA, guideline-grounded QA derived from 46,469 clinical practice guidelines, and clinical vignettes. The pipeline enforces system-wide decontamination, gold-label resampling of teacher generations, and end-to-end validation by a four-physician panel. We evaluate using an LLM-as-a-judge protocol over expert-written clinical vignettes, calibrated against 204 human raters. We apply the recipe to five FO base models (Apertus-70B/8B-Instruct, OLMo-2-32B-SFT, EuroLLM-22B/9B-Instruct). All MeditronFO variants are preferred over their bases. Apertus-70B-MeditronFO improves +6.6 points over its base (47.2% to 53.8%) on aggregate medical benchmarks, establishing a new FO SoTA. Gemma-3-27B-MeditronFO is preferred over MedGemma in 58.6% of LLM-as-a-judge comparisons and outperforms it on HealthBench (58% vs 55.9%). These results show that fully open pipelines can achieve state-of-the-art domain-specific performance without sacrificing auditability or reproducibility.

Lars Klein, Nearchos Potamitis, Roland Aydin et al.

While numerous frameworks have been developed to enhance the reasoning abilities of large language models (LLMs), there is a scarcity of methods that effectively balance the trade-off between cost and quality. In this paper, we introduce Fleet of Agents (FoA), a novel and intuitive yet principled framework utilizing LLMs as agents to navigate through dynamic tree searches, employing a genetic-type particle filtering approach. FoA spawns a multitude of agents, each exploring the search space autonomously, followed by a selection phase where resampling based on a heuristic value function optimizes the balance between exploration and exploitation. This mechanism enables dynamic branching, adapting the exploration strategy based on discovered solutions. We conduct extensive experiments on three benchmark tasks, ``Game of 24'', ``Mini-Crosswords'', and ``WebShop'', utilizing four different LLMs, ``GPT-3.5'', ``GPT-4'', ``LLaMA3.2-11B'', and ``LLaMA3.2-90B''. On average across all tasks and LLMs, FoA obtains a quality improvement of ~5% while requiring only ~40% of the cost of previous SOTA methods. Notably, our analyses reveal that (1) FoA achieves the best cost-quality trade-off among all benchmarked methods and (2) FoA + LLaMA3.2-11B surpasses the Llama3.2-90B model. FoA is publicly available at https://github.com/au-clan/FoA.

David Schulmeister, Valentin Hartmann, Lars Klein et al.

Today, a lot of research on language models is focused on large, general-purpose models. However, many NLP pipelines only require models with a well-defined, small set of capabilities. While large models are capable of performing the tasks of those smaller models, they are simply not fast enough to process large amounts of data or offer real-time responses. Furthermore, they often use unnecessarily large amounts of energy, leading to sustainability concerns and problems when deploying them on battery-powered devices. In our work, we show how to train small models for such efficiency-critical applications. As opposed to many off-the-shelf NLP pipelines, our models use modern training techniques such as distillation, and offer support for low-resource languages. We call our models TiME (Tiny Monolingual Encoders) and comprehensively evaluate them on a range of common NLP tasks, observing an improved trade-off between benchmark performance on one hand, and throughput, latency and energy consumption on the other. Along the way, we show that distilling monolingual models from multilingual teachers is possible, and likewise distilling models with absolute positional embeddings from teachers with relative positional embeddings.

Bardia Mohammadi, Nearchos Potamitis, Lars Klein et al.

LLM agents increasingly act on external systems, yet tool effects are immediate. Under failures, speculation, or contention, losing branches can leak unintended side effects with no safe rollback. We introduce Atomix, a runtime that provides progress-aware transactional semantics for agent tool calls. Atomix tags each call with an epoch, tracks per-resource frontiers, and commits only when progress predicates indicate safety; bufferable effects can be delayed, while externalized effects are tracked and compensated on abort. Across real workloads with fault injection, transactional retry improves task success, while frontier-gated commit strengthens isolation under speculation and contention.

Roland Aydin, Lars Klein, Arnaud Miribel et al.

The learning of a new language remains to this date a cognitive task that requires considerable diligence and willpower, recent advances and tools notwithstanding. In this paper, we propose Broccoli, a new paradigm aimed at reducing the required effort by seamlessly embedding vocabulary learning into users' everyday information diets. This is achieved by inconspicuously switching chosen words encountered by the user for their translation in the target language. Thus, by seeing words in context, the user can assimilate new vocabulary without much conscious effort. We validate our approach in a careful user study, finding that the efficacy of the lightweight Broccoli approach is competitive with traditional, memorization-based vocabulary learning. The low cognitive overhead is manifested in a pronounced decrease in learners' usage of mnemonic learning strategies, as compared to traditional learning. Finally, we establish that language patterns in typical information diets are compatible with spaced-repetition strategies, thus enabling an efficient use of the Broccoli paradigm. Overall, our work establishes the feasibility of a novel and powerful "install-and-forget" approach for embedded language acquisition.