24 papers
Verbalizing Activation Differences with Natural Language Autoencoders
Andrea Blasi Núñez, Gianluca Barmina, Federico Torrielli +3
Activation verbalizers translate the internal representations of large language models into human-readable text. The recently proposed natural language autoencoders (NLAs) learn th…
Multilingual GSM-Symbolic: What determines capability transfer across languages?
Kenneth Enevoldsen, Riley Herchert, Sofie Mosegaard +22
We understand little about how capabilities acquired in one language carry over to another, or what governs this transfer: evaluations rely on incomparable, saturation-prone datase…
Selecting The Most Informative Tokens in Natural Language Autoencoders
Federico Torrielli, Gianluca Barmina, Andrea Blasi Núñez +4
Natural language autoencoders translate a language model's internal activations into readable explanations. Explaining every token position is costly. Which positions should an aud…
Stale-Document Poisoning: When Outdated Retrieval Overrides Correct Model Answers
Md Shamim Ahmed, Lukas Galke Poech, Richard Röttger
Retrieval-augmented generation (RAG) is often used to address outdated knowledge by providing external evidence. But retrieval helps only when that evidence is still valid. We iden…
Rethinking Pivot Programming Languages in Code Language Models
Andor Diera, Lukas Galke Poech, Matthias Tichy
Multilingual code language models transfer skills across programming languages (PLs), but whether any PL occupies a privileged pivot position remains contested: geometric analyses…
Write Once, Run Everywhere: The Axon DSL for Shape-Safe and Framework-Agnostic LLM Architectures
Jacob Nielsen, Danial Namazifard, Lukas Galke Poech +1
The entire ecosystem of open-source language models effectively relies on a single platform. What if this platform was forced to shut down tomorrow? Implementing and maintaining ef…
Mimir: A Sub-2B Hierarchical Reasoning Model Trained on Permissible Data Punches Above Its Weight Class
Peter Schneider-Kamp, Jacob Nielsen, Gianluca Barmina +4
Training capable language models typically relies on web-scale pretraining followed by multi-stage post-training, creating substantial computational and data-governance barriers fo…
The Energy Society: A Simulation Environment for Studying Agent Cooperation under Survival Pressure
Lucas Bergholdt Hansen, Federico Torrielli, Filippo Tonini +1
LLM-based agents are increasingly deployed in multi-agent environments whose incentives can shape their behavior. We introduce The Energy Society, a minimal survival economy for st…
Influence of Prompt Engineering on Small Language Models for Guarded Query Routing
Richard Šléher, William Brach, Kristián Košťál +1
We study the problem of guarded query routing, where we assume that a user query first meets a router that either determines the ideal endpoint for in-distribution queries or rejec…
The Arbiter Agent: Continually Monitoring Multi-Agent Conversations to Detect Emergent Misalignment
Filippo Tonini, Federico Torrielli, Anton Danholt Lautrup +3
As AI systems built from multiple language-model agents become more common, they are increasingly used to make decisions together: discussing, negotiating, and acting on shared tas…
BrainSurgery: Reproducible and Reliable Declarative Weight Manipulations for Model Editing and Upcycling
Gianluca Barmina, Annemette Broch Pirchert, Andrea Blasi Núñez +2
As deep learning models scale, managing, inspecting, and modifying large checkpoints has become increasingly challenging. Researchers often need to alter model weights for layer re…
PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models
Gianluca Barmina, Federico Torrielli, Sven Harms +7
Large language models (LLMs) routinely face requests that should be refused, creating a trade-off between helpfulness and harm prevention. However, refusals themselves can be helpf…
LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMs
Gianluca Barmina, Peter Schneider-Kamp, Lukas Galke Poech
Large language models can reproduce training data, but existing memorization evaluations mostly measure whether models can be forced to do so, rather than whether they do so under…
Emergent Languages in Populations of Language Model Agents: From Token Efficiency to Oversight Evasion
Stine Lyngsø Beltoft, William Brach, Federico Torrielli +5
Monitoring autonomous language model agents currently relies mostly on surface behavior. But what happens when agent populations invent new languages with the goal of avoiding huma…
ChronoMedKG: A Temporally-Grounded Biomedical Knowledge Graph and Benchmark for Clinical Reasoning
Md Shamim Ahmed, Farzaneh Firoozbakht, Lukas Galke Poech +2
Biomedical knowledge graphs (KGs) treat disease associations as static facts, but temporal information is crucial for clinical reasoning, e.g., a symptom diagnostic of one disease…
The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment
William Brach, Federico Torrielli, Stine Lyngsø Beltoft +3
Moltbook is a Reddit-like platform where OpenClaw agents post, comment, and vote at scale - a so far unprecedented incident that comes with serious safety concerns. With the aim of…
The Provenance Gap in Clinical AI: Evidence-Traceable Temporal Knowledge Graphs for Rare Disease Reasoning
Md Shamim Ahmed, Maja Dusanic, Moritz Nikolai Kirschner +4
Frontier large language models generate clinically accurate outputs, but their citations are often fabricated. We term this the Provenance Gap. We tested five frontier LLMs across…
Confidence and Calibration of Activation Oracles for Reliable Interpretation of Language Model Internals
Federico Torrielli, Peter Schneider-Kamp, Lukas Galke Poech
An activation oracle is a language model trained to convert the activations of another model into natural language. However, it is so far unclear to what extent these verbalization…
SommBench: Assessing Sommelier Expertise of Language Models
William Brach, Tomas Bedej, Jacob Nielsen +10
With the rapid advances of large language models, it becomes increasingly important to systematically evaluate their multilingual and multicultural capabilities. Previous cultural…
FlexMoRE: A Flexible Mixture of Rank-heterogeneous Experts for Efficient Federatedly-trained Large Language Models
Annemette Brok Pirchert, Jacob Nielsen, Mogens Henrik From +2
Recent advances in mixture-of-experts architectures have shown that individual experts models can be trained federatedly, i.e., in isolation from other experts by using a common ba…
SDUs DAISY: A Benchmark for Danish Culture
Jacob Nielsen, Stine L. Beltoft, Peter Schneider-Kamp +1
We introduce Daisy, a factual knowledge benchmark for Danish cultural heritage, based on curated topics from the Danish Culture Canon 2006. For each artifact in the culture canon,…
DaLA: Danish Linguistic Acceptability Evaluation Guided by Real World Errors
Gianluca Barmina, Nathalie Carmen Hau Norman, Peter Schneider-Kamp +1
We present an enhanced benchmark for evaluating linguistic acceptability in Danish. We first analyze the most common errors found in written Danish. Based on this analysis, we intr…
Training Language Models to Use Prolog as a Tool
Niklas Mellgren, Peter Schneider-Kamp, Lukas Galke Poech
Language models frequently produce plausible yet incorrect reasoning traces that are difficult to verify. We investigate fine-tuning models to use Prolog as an external symbolic re…
Chain of Summaries: Summarization Through Iterative Questioning
William Brach, Kristián Košťál, Lukas Galke Poech
Large Language Models (LLMs) are increasingly using external web content. However, much of this content is not easily digestible by LLMs due to LLM-unfriendly formats and limitatio…