arXiv subjects

Joachim Baumann

Publications and source records attributed to Joachim Baumann.

At least 19 recordsLinked to original sources

arXiv2026

API Benchmark Scores Do Not Reliably Transfer to Chatbot Interfaces

Benchmark scores are a central currency in model releases: they inform purchasing decisions, shape public trust, and influence policy. Yet, a key assumption underlying benchmark scores is that the model performance measured through APIs faithfully reflects the behavior of deployed systems. We challenge this assumption by auditing ChatGPT, Claude, and Gemini across seven systems and nine benchmarks spanning general capability, social bias, and sycophancy. We find systematic API--interface differences in both accuracy and consistency. On average, API evaluations score 3.4 percentage points higher in accuracy and 2.1 percentage points higher in test--retest agreement than corresponding interface evaluations. For ChatGPT, the performance difference between API and interface access rivals the API-only difference between GPT 5.3 and GPT 5.4. Put differently, switching access surfaces can degrade performance as much as downgrading a full model generation. We further test whether exposed API controls can reproduce interface behavior by varying system prompts, sampling parameters, and reasoning settings. These controls shift behavior in some cases but do not reliably eliminate the gap. Our findings document a context-validity gap: measurements obtained through APIs do not necessarily generalize to corresponding deployed interfaces, complicating the use of API evaluations as proxies for deployed systems.

cs.AI↗

arXiv2026

On the Societal Impact of Machine Learning

This PhD thesis investigates the societal impact of machine learning (ML). ML increasingly informs consequential decisions and recommendations, significantly affecting many aspects of our lives. As these data-driven systems are often developed without explicit fairness considerations, they carry the risk of discriminatory effects. The contributions in this thesis enable more appropriate measurement of fairness in ML systems, systematic decomposition of ML systems to anticipate bias dynamics, and effective interventions that reduce algorithmic discrimination while maintaining system utility. I conclude by discussing ongoing challenges and future research directions as ML systems, including generative artificial intelligence, become increasingly integrated into society. This work offers a foundation for ensuring that ML's societal impact aligns with broader social values.

cs.LG↗

arXiv2026

Prompt Revision as a Source of Cultural Bias in Text-to-Image Systems

Commercial text-to-image systems silently revise user prompts before generating images, a step users typically cannot disable or even see. Yet, existing audits of cultural bias examine only the final images and treat generation as a single pipeline, so they cannot tell where the bias originates. We introduce WORLDVIEW, a multilingual benchmark of 8,960 prompts across 15 languages and 31 language-context pairings. Using it, we audit the revision layer in three systems (DALL-E-3, Imagen-4, GPT-Image-1.5) through a three-step analysis of how heavily it marks each cultural context, whether it flattens that context into a narrow vocabulary, and whether that vocabulary is stereotypical. Relative to a no-context English baseline, the US is the least-marked context, while non-Western and non-Anglophone contexts are marked far more heavily, flattened into narrow vocabularies applied across topically diverse prompts, and reduced to recognizable cultural stereotypes. Comparing images from original versus revised prompts on models without a revision layer, we identify the layer itself as a previously undocumented, causal source of this stereotyping. To locate cultural bias, and fix it, we must audit the system as deployed, not the model alone.

cs.AI↗

arXiv2026

WildSEEK: Evaluating Language Models for Information-Seeking

Language models are increasingly mediating information access to end users, urging a systematic evaluation of their responses for a fair and reliable information ecosystem. Existing evaluations, however, are often topic-specific or synthetic, limiting their ability to capture the complexity of "in the wild" information-seeking queries and the risks present in model responses. To address this gap, we introduce WildSEEK, a manually annotated dataset of 3k information-seeking queries from real user interactions, and an evaluation framework for LLM-generated responses. WildSEEK includes annotations for risk-sensitive domains (e.g. health and financial information), and distinguishes factoid queries from analytical queries which seek responses beyond facts. We train classifiers on WildSEEK to analyze more than 1.8M realistic user queries. We find that over a third of information-seeking queries are high-risk and more often analytical. Our findings show that LLM responses fail more often in four criteria: sycophantic behavior, overreliance, a default US-centric perspective, and poor handling of vulnerable populations -- with failure rates being mostly higher for analytical queries. By providing methods to monitor the reliability, safety, and fairness of LLM behavior, our dataset and evaluation framework offer an empirical foundation for the broader question of how these systems should behave as they take on a growing role in information access.

cs.CL↗

arXiv2026

Humanly: A Configurable and Traceable Environment for Human-AI Collaborative Writing

Teachers, conference chairs, and public readers all judge writing from limited evidence, seeing only a finished document and not the process that produced it. Final text alone cannot reveal whether a document was produced through human typing, AI generation, or mixed human-AI collaboration. Existing process-tracking tools help, but many are tied to host-document histories, provide coarse activity records, and offer limited control over the writing environment. Humanly is a writing platform that makes the writing process itself the evidence. Users configure writing environments for personal documents or assigned tasks and draft in a workspace that records writing activity and in-platform AI assistance. Humanly can package a completed session into a sealed writing certificate with configuration-aware anomaly behavior review. It can support writing scenarios such as course assignments, peer review, and personal certification. Our user study shows that Humanly is helpful across roles, and a red-teaming study shows that the Humanly Typing Detector distinguishes human hand typing from automated typing.

cs.CL↗

arXiv2026

Pluralis v0.1: Towards a Multicultural, Multimodal, Multilingual Benchmark for AI Risk and Reliability

Current AI safety evaluation and benchmarking frameworks predominantly rely on Western-centric culture-agnostic defaults that mask critical regional laws, socio-linguistic nuances, and cultural taboos, leaving Vision-Language Models (VLMs) vulnerable in global deployments. We introduce Pluralis v0.1: a novel multimodal, multi-regional, and multilingual dataset built from a culture-first perspective. Spanning 6,448 prompts across six Asia-Pacific countries (Bangladesh, India, Korea, Pakistan, Singapore, Taiwan) and eight languages, Pluralis diverges from prior work by natively sourcing localized safety hazards rather than adapting Western datasets. Crucially, it introduces a multimodal evaluation paradigm: user text (e.g., "Should I gift this?") and an image referring to "this" (e.g., a clock) - both innocuous in isolation, but synergistically triggering specific legal or cultural violations. Pluralis disentangles universal safety violations from localized cultural appropriateness, establishing the latter as a first-class evaluation axis. To operationalize this, we present Judge-Pluralis, an agreement-gated LLM-as-a-Judge ensemble trained on examples classified in an empirically derived cultural taxonomy. Observing VLM behavior on a subset of the Pluralis surfaces recurring, locale-specific failure modes such as image misidentifications with downstream harm, missed item-context-locale interactions, and inadequate refusals. These failure modes vary systematically across locales and languages, exposing blind spots that globally averaged metrics conceal. Ultimately, Pluralis is not presented as a solved evaluation framework for cultural alignment, but rather as a first step and catalyst for future innovation. We call upon the research community to utilize this foundation to advance the science of multilingual, multicultural evaluation to better support AI cultural alignment globally.

cs.CL↗

arXiv2026

Stop Automating Peer Review Without Rigorous Evaluation

Large language models offer a tempting solution to address the peer review crisis. This position paper argues that today's AI systems should not be used to produce paper reviews. We ground this position in an empirical comparison of human- versus AI-generated ICLR 2026 reviews and an evaluation of the effect of automated paper rewriting on different AI reviewers. We identify two critical issues: 1) AI reviewers exhibit a hivemind effect of excessive agreement within and across papers that reduces perspective diversity. 2) AI review scores are trivially gameable through paper laundering: prompting an LLM to rewrite a paper could significantly increase the scores from AI reviewers, demonstrating that LLM reviewers are easy to game through stylistic changes rather than scientific results. However, non-gameability and review diversity are necessary but not sufficient conditions for automation. We argue that addressing the peer review crisis requires a science of peer review automation -- not general-purpose LLMs deployed without rigorous evaluation.

cs.AI↗

arXiv2026

SecureForge: Finding and Preventing Vulnerabilities in LLM-Generated Code via Prompt Optimization

LLM coding agents now generate code at an unprecedented scale, yet LLM-generated code introduces cybersecurity vulnerabilities into codebases without human involvement. Even when frontier models are explicitly asked to write secure production code with relevant weaknesses to avoid in context, we find that they still produce verifiable vulnerabilities on average 23% of the time across a corpus of 250 benign coding prompts. We introduce SecureForge, an automated pipeline that both audits security risks of frontier models and produces auditing-informed secure system prompts that reduce output security vulnerabilities while maintaining unit test performance. SecureForge first identifies benign prompts that produce statically detectable vulnerabilities, and then amplifies them into a large synthetic prompt corpus of diverse scenarios using a Markovian sampling technique to jointly maintain error rates and prompt diversity. This corpus is then used to iteratively optimize the system prompts to reduce output security vulnerabilities. On frontier models, SecureForge yields a statistically significant Pareto improvement in both unit test success and output security, with output vulnerabilities reduced by up to 48%. The resulting system prompts transfer zero-shot to in-the-wild coding agent prompts, without any exposure to real user prompt distributions during optimization.

cs.CR↗

arXiv2026

SWE-chat: Coding Agent Interactions From Real Users in the Wild

AI coding agents are being adopted at scale, yet we lack empirical evidence on how people actually use them and how much of their output is useful in practice. We present SWE-chat, the first large-scale dataset of real coding agent sessions collected from open-source developers in the wild. The dataset currently contains 6,000 sessions, comprising more than 63,000 user prompts and 355,000 agent tool calls. SWE-chat is a living dataset; our collection pipeline automatically and continually discovers and processes sessions from public repositories. Leveraging SWE-chat, we provide an initial empirical characterization of real-world coding agent usage and failure modes. We find that coding patterns are bimodal: in 41% of sessions, agents author virtually all committed code ("vibe coding"), while in 23%, humans write all code themselves. Despite rapidly improving capabilities, coding agents remain inefficient in natural settings. Just 44% of all agent-produced code survives into user commits, and agent-written code introduces more security vulnerabilities than code authored by humans. Furthermore, users push back against agent outputs -- through corrections, failure reports, and interruptions -- in 44% of all turns. By capturing complete interaction traces with human vs. agent code authorship attribution, SWE-chat provides an empirical foundation for moving beyond curated benchmarks towards an evidence-based understanding of how AI agents perform in real developer workflows.

cs.AI↗

arXiv2026

SimBench: Benchmarking the Ability of Large Language Models to Simulate Human Behaviors

Large language model (LLM) simulations of human behavior have the potential to revolutionize the social and behavioral sciences, if and only if they faithfully reflect real human behaviors. Current evaluations of simulation fidelity are fragmented, based on bespoke tasks and metrics, creating a patchwork of incomparable results. To address this, we introduce SimBench, the first large-scale, standardized benchmark for a robust, reproducible science of LLM simulation. By unifying 20 diverse datasets covering tasks from moral decision-making to economic choice across a large global participant pool, SimBench provides the necessary foundation to ask fundamental questions about when, how, and why LLM simulations succeed or fail. We show that the best LLMs today achieve meaningful but modest simulation fidelity (score: 40.80/100), with performance scaling log-linearly with model size but not with increased inference-time compute. We discover an alignment-simulation tradeoff: instruction tuning improves performance on low-entropy (consensus) questions but degrades it on high-entropy (diverse) ones. Models particularly struggle when simulating specific demographic groups. Finally, we demonstrate that simulation ability correlates most strongly with knowledge-intensive reasoning (MMLU-Pro, r = 0.939). By making progress measurable, we aim to accelerate the development of more faithful LLM simulators.

cs.CL↗

arXiv2026

Auditing Google's AI Overviews and Featured Snippets: A Case Study on Baby Care and Pregnancy

Google Search increasingly surfaces AI-generated content through features like AI Overviews (AIO) and Featured Snippets (FS), which users frequently rely on despite having no control over their presentation. Through a systematic algorithm audit of 1,508 real baby care and pregnancy-related queries, we evaluate the quality and consistency of these information displays. Our robust evaluation framework assesses multiple quality dimensions, including answer consistency, relevance, presence of medical safeguards, source categories, and sentiment alignment. Our results reveal concerning gaps in information consistency, with information in AIO and FS displayed on the same search result page being inconsistent with each other in 33% of cases. Despite high relevance scores, both features critically lack medical safeguards (present in just 11% of AIO and 7% of FS responses). While health and wellness websites dominate source categories for both, AIO and FS, FS also often link to commercial sources. These findings have important implications for public health information access and demonstrate the need for stronger quality controls in AI-mediated health information. Our methodology provides a transferable framework for auditing AI systems across high-stakes domains where information quality directly impacts user well-being.

cs.CL↗

arXiv2026

GoogleTrendArchive: A Year-Long Archive of Real-Time Web Search Trends Worldwide

GoogleTrendArchive is a comprehensive archive of Google Trending Now data spanning over one year (from November 28, 2024 to January 3, 2026) across 125 countries and 1,358 locations. Unlike Google Trends, which requires specifying search terms in advance, Trending Now captures search queries experiencing real-time surges, offering a way to inductively discover trending patterns across regions for studying collective attention dynamics. However, Google does not provide historical access to this data beyond seven days. Our dataset addresses this gap by presenting an archive of Trending Now data. The dataset contains over 7.6 million trend episodes. Each record includes the trend identifier, search volume bucket, precise timestamps, duration, geographic location, and related query clusters. This dataset, among other, enables systematic studies of information diffusion patterns, cross-cultural attention dynamics, crisis responses, and the temporal evolution of collective information-seeking at a global scale. The comprehensive geographic coverage facilitates fine-grained cross-country or cross-regional comparative analyses.

cs.IR↗

arXiv2025

Reduced AI Acceptance After the Generative AI Boom: Evidence From a Two-Wave Survey Study

The rapid adoption of generative artificial intelligence (GenAI) technologies has led many organizations to integrate AI into their products and services, often without considering user preferences. Yet, public attitudes toward AI use, especially in impactful decision-making scenarios, are underexplored. Using a large-scale two-wave survey study (n_wave1=1514, n_wave2=1488) representative of the Swiss population, we examine shifts in public attitudes toward AI before and after the launch of ChatGPT. We find that the GenAI boom is significantly associated with reduced public acceptance of AI (see Figure 1) and increased demand for human oversight in various decision-making contexts. The proportion of respondents finding AI "not acceptable at all" increased from 23% to 30%, while support for human-only decision-making rose from 18% to 26%. These shifts have amplified existing social inequalities in terms of widened educational, linguistic, and gender gaps post-boom. Our findings challenge industry assumptions about public readiness for AI deployment and highlight the critical importance of aligning technological development with evolving public preferences.

cs.AI↗

arXiv2025

Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation

Large language models are rapidly transforming social science research by enabling the automation of labor-intensive tasks like data annotation and text analysis. However, LLM outputs vary significantly depending on the implementation choices made by researchers (e.g., model selection or prompting strategy). Such variation can introduce systematic biases and random errors, which propagate to downstream analyses and cause Type I (false positive), Type II (false negative), Type S (wrong sign), or Type M (exaggerated effect) errors. We call this phenomenon where configuration choices lead to incorrect conclusions LLM hacking. We find that intentional LLM hacking is strikingly simple. By replicating 37 data annotation tasks from 21 published social science studies, we show that, with just a handful of prompt paraphrases, virtually anything can be presented as statistically significant. Beyond intentional manipulation, our analysis of 13 million labels from 18 different LLMs across 2361 realistic hypotheses shows that there is also a high risk of accidental LLM hacking, even when following standard research practices. We find incorrect conclusions in approximately 31% of hypotheses for state-of-the-art LLMs, and in half the hypotheses for smaller language models. While higher task performance and stronger general model capabilities reduce LLM hacking risk, even highly accurate models remain susceptible. The risk of LLM hacking decreases as effect sizes increase, indicating the need for more rigorous verification of LLM-based findings near significance thresholds. We analyze 21 mitigation techniques and find that human annotations provide crucial protection against false positives. Common regression estimator correction techniques can restore valid inference but trade off Type I vs. Type II errors. We publish a list of practical recommendations to prevent LLM hacking.

cs.CL↗

arXiv2025

AILuminate: Introducing v1.0 of the AI Risk and Reliability Benchmark from MLCommons

The rapid advancement and deployment of AI systems have created an urgent need for standard safety-evaluation frameworks. This paper introduces AILuminate v1.0, the first comprehensive industry-standard benchmark for assessing AI-product risk and reliability. Its development employed an open process that included participants from multiple fields. The benchmark evaluates an AI system's resistance to prompts designed to elicit dangerous, illegal, or undesirable behavior in 12 hazard categories, including violent crimes, nonviolent crimes, sex-related crimes, child sexual exploitation, indiscriminate weapons, suicide and self-harm, intellectual property, privacy, defamation, hate, sexual content, and specialized advice (election, financial, health, legal). Our method incorporates a complete assessment standard, extensive prompt datasets, a novel evaluation framework, a grading and reporting system, and the technical as well as organizational infrastructure for long-term support and evolution. In particular, the benchmark employs an understandable five-tier grading scale (Poor to Excellent) and incorporates an innovative entropy-based system-response evaluation. In addition to unveiling the benchmark, this report also identifies limitations of our method and of building safety benchmarks generally, including evaluator uncertainty and the constraints of single-turn interactions. This work represents a crucial step toward establishing global standards for AI risk and reliability evaluation while acknowledging the need for continued development in areas such as multiturn interactions, multimodal understanding, coverage of additional languages, and emerging hazard categories. Our findings provide valuable insights for model developers, system integrators, and policymakers working to promote safer AI deployment.

cs.CY↗

arXiv2025

Algorithmic Collective Action in Recommender Systems: Promoting Songs by Reordering Playlists

We investigate algorithmic collective action in transformer-based recommender systems. Our use case is a music streaming platform where a collective of fans aims to promote the visibility of an underrepresented artist by strategically placing one of their songs in the existing playlists they control. We introduce two easily implementable strategies to select the position at which to insert the song with the goal to boost recommendations at test time. The strategies exploit statistical properties of the learner by targeting discontinuities in the recommendations, and leveraging the long-tail nature of song distributions. We evaluate the efficacy of our strategies using a publicly available recommender system model released by a major music streaming platform. Our findings reveal that through strategic placement even small collectives (controlling less than 0.01\% of the training data) can achieve up to $40\times$ more test time recommendations than an average song with the same number of training set occurrences. Focusing on the externalities of the strategy, we find that the recommendations of other songs are largely preserved, and the newly gained recommendations are distributed across various artists. Together, our findings demonstrate how carefully designed collective action strategies can be effective while not necessarily being adversarial.

cs.IR↗

arXiv2024

Preventing Eviction-Caused Homelessness through ML-Informed Distribution of Rental Assistance

Rental assistance programs provide individuals with financial assistance to prevent housing instabilities caused by evictions and avert homelessness. Since these programs operate under resource constraints, they must decide who to prioritize. Typically, funding is distributed by a reactive or first-come-first serve allocation process that does not systematically consider risk of future homelessness. We partnered with Allegheny County, PA to explore a proactive allocation approach that prioritizes individuals facing eviction based on their risk of future homelessness. Our ML system that uses state and county administrative data to accurately identify individuals in need of support outperforms simpler prioritization approaches by at least 20% while being fair and equitable across race and gender. Furthermore, our approach would identify 28% of individuals who are overlooked by the current process and end up homeless. Beyond improvements to the rental assistance program in Allegheny County, this study can inform the development of evidence-based decision support tools in similar contexts, including lessons about data needs, model design, evaluation, and field validation.

cs.CY↗

arXiv2023

On Prediction-Modelers and Decision-Makers: Why Fairness Requires More Than a Fair Prediction Model

An implicit ambiguity in the field of prediction-based decision-making regards the relation between the concepts of prediction and decision. Much of the literature in the field tends to blur the boundaries between the two concepts and often simply speaks of 'fair prediction.' In this paper, we point out that a differentiation of these concepts is helpful when implementing algorithmic fairness. Even if fairness properties are related to the features of the used prediction model, what is more properly called 'fair' or 'unfair' is a decision system, not a prediction model. This is because fairness is about the consequences on human lives, created by a decision, not by a prediction. We clarify the distinction between the concepts of prediction and decision and show the different ways in which these two elements influence the final fairness properties of a prediction-based decision system. In addition to exploring this relationship conceptually and practically, we propose a framework that enables a better understanding and reasoning of the conceptual logic of creating fairness in prediction-based decision-making. In our framework, we specify different roles, namely the 'prediction-modeler' and the 'decision-maker,' and the information required from each of them for being able to implement fairness of the system. Our framework allows for deriving distinct responsibilities for both roles and discussing some insights related to ethical and legal requirements. Our contribution is twofold. First, we shift the focus from abstract algorithmic fairness to context-dependent decision-making, recognizing diverse actors with unique objectives and independent actions. Second, we provide a conceptual framework that can help structure prediction-based decision problems with respect to fairness issues, identify responsibilities, and implement fairness governance mechanisms in real-world scenarios.

cs.CY↗