Searcharxiv⌕ Search

arXiv subjects

Gerrit Großmann

Publications and source records attributed to Gerrit Großmann.

At least 19 recordsLinked to original sources

AI Alignment Is Neither Solvable nor Sufficient

Humans and AI systems increasingly act within the same social and economic environments. Alignment, the goal of ensuring that AI systems act in accordance with human values, is widely regarded as a central challenge for ensuring that AI benefits society while avoiding harm. We argue that this endeavor is misguided. First, alignment is not well defined, as it presumes a sufficiently stable and identifiable set of human intentions or values. Second, where a concrete notion of alignment can be specified, it is not solvable. Third, where alignment appears solvable, solving it would not necessarily mitigate the negative consequences of AI. A key reason is that alignment is typically framed as a property of individual AI systems, while the resulting hybrid societies of humans and machines are complex systems shaped by interactions and feedback loops. Emergent risks, and not only those posed by individual misaligned systems, are therefore likely to become a major source of harm. We conclude that AI safety is an ongoing challenge and not a problem that can be solved once and for all by verifying a particular model property.

cs.AI↗

Seahorse: A Unified Benchmarking Framework for Spatiotemporal Event Modeling

Spatiotemporal point processes (STPPs) model event data in continuous time and space, with applications in mobility, epidemiology, and public safety. Recent neural STPPs span expressive intensities, density models, continuous-time latent dynamics, normalizing-flow spatial decoders, and score-based generative mechanisms. Yet comparison remains fragile because implementations differ in preprocessing, coordinate normalization, splits, likelihood conventions, and evaluation protocols. We present Seahorse, a unified framework for reproducible training and evaluation of STPPs. Seahorse provides an encode-evolve-decode interface and an executable benchmarking contract that standardizes dataset handling, tuning, raw-coordinate likelihood reporting, and run artifacts across classical baselines, neural likelihood models, continuous-time Neural STPP variants, and sample-based generative models. On 13 real datasets and 19 presets, no model dominates: likelihood, calibration, and next-event prediction favor different models, and the strongest models take from minutes to hours to train. Using HawkesNest as a controlled synthetic stress-test suite, we show how diagnostic benchmarking can probe neural STPP behavior under increasing spatiotemporal entanglement. Seahorse provides reusable infrastructure for comparing, diagnosing, and extending STPP models beyond implementation-specific leaderboards. The implementation, including preset configurations and reproduction scripts, is available at https://github.com/YahyaAalaila/seahorse.

cs.LG↗

Rewarded for Not Getting Caught: Why Alignment Training Selects for Conditional Compliance and Cannot Instill Norms

Current alignment training cannot be expected to deliver what it is meant to deliver: AI agents that comply with norms whatever the gains and whether or not anyone is watching. Training that rewards and penalizes behavior can teach norms only through that reward, and a reward can encode a norm only as a price: "do not do X" becomes "doing X costs something", and training charges the cost only when X is noticed. A price, unlike a norm, can be paid: a large enough gain outweighs it, noticed or not. And because the cost is charged only where violations are noticed, training cannot favor a policy that always complies over one that complies only where it would be caught. When an unnoticed violation is the only way to complete a task, it even favors the latter, the more so the more strongly it rewards task success. Agency makes this the normal case: agents mostly act without oversight, can often tell when they are observed, and may take steps to evade observation. That agents behave as intended when they infer they are being tested, and differently when they infer they are not, is therefore no malfunction; the training regime is working as built. This reading unifies alignment faking, sandbagging, and evaluation-aware scheming. No behavioral evaluation can show that a model does better, since that would require observing how it behaves unobserved; a model that complies when it seems unobserved may simply have recognized the test. Easing the pressure on agents to complete tasks at any cost can make violations rarer, but never impossible. For norms that no gain should be allowed to outweigh, the remedy is not deeper internalization but architecture that makes violations unavailable rather than unchosen. We state this argument explicitly, premise by premise, read recent findings and a real-world incident against it, and draw consequences for design and governance.

cs.AI↗

Multi-Scale Spatio-Temporal Exploratory Analysis of Urban Vehicle Theft

Vehicle theft is a persistent urban problem with large economic and social costs. We analyze 521,098 stolen-vehicle reports filed with the Los Angeles Police Department from 2000 to 2022, at scales from hours to decades and from 100 m grid cells to census tracts. Much of the work lies in making the administrative records fit for analysis: we separate thefts from recoveries and traffic reports filed in the same records, repair time fields, validate the geocoding, and harmonize census data across changing tract boundaries. Theft is strongly and persistently concentrated: 10.8% of 250 m cells hold half of all thefts, and six census tracts are significant rate hotspots in all 23 years after false discovery rate control. Thefts close in space and time are more frequent than expected by chance, but about half of this excess comes from events at identical coordinates. Lower-income tracts have more thefts at equal exposure, monthly temperature shows no association with theft, and thefts peak in the evening. From 2015-2019 to 2021-2022, the share of Kia and Hyundai among stolen vehicles rose from 1.5% to 9.5% and from 1.6% to 7.7%, mostly among model years 2011-2021, which largely lacked a standard immobilizer. All findings are descriptive. We also describe how the data quality issues we corrected would otherwise distort several of these patterns.

cs.CY↗

Decomposing Visual Classification: Assessing Tree-Based Reasoning in VLMs

Vision language models (VLMs) excel at zero-shot visual classification, but their performance on fine-grained tasks and large hierarchical label spaces is understudied. This paper investigates whether structured, tree-based reasoning can enhance VLM performance. We introduce a framework that decomposes classification into interpretable decisions using decision trees and evaluates it on fine-grained (GTSRB) and coarse-grained (CIFAR-10) datasets. Although the model achieves 98.2% accuracy in understanding the tree knowledge, tree-based reasoning consistently underperforms standard zero-shot prompting. We also explore enhancing the tree prompts with LLM-generated classes and image descriptions to improve alignment. The added description enhances the performance of the tree-based and zero-shot methods. Our findings highlight limitations of structured reasoning in visual classification and offer insights for designing more interpretable VLM systems.

cs.CV↗

MYOPIA: Understanding the Limits of Tree-Based Reasoning in Vision-Language Models

Decision trees can make vision-language model (VLM) classification interpretable by replacing a single prediction with a sequence of root-to-leaf questions. However, the accuracy cost of this greedy, non-backtracking procedure remains unclear. We introduce \textbf{MYOPIA}, a benchmark comprising four LLM-generated, human-refined decision trees for GTSRB, MNIST, CIFAR-10, and MSCOCO, together with matched prompts, diagnostic protocols, and an evaluation harness. Across GPT-4o, Qwen-VL Max, and LLaMA-3.2 11B Vision, tree traversal underperforms zero-shot prompting in all 12 dataset--model pairs; paired bootstrap analyses consistently favor zero-shot. Our diagnostics characterize interacting failure modes associated with this gap. Models often recall tree-encoded class attributes more accurately than they ground the same attributes in images: the mean label--image accuracy gaps are 9.5, 13.8, and 12.6 points for GPT-4o, Qwen-VL Max, and LLaMA, respectively. Incorrect local decisions then irreversibly exclude the correct class, while answer extraction introduces additional model-dependent loss. A node-level audit of the four trees finds that split entropy is weakly but consistently associated with lower node accuracy, whereas depth, branching factor, and within-node grounding differences are less stable. \textbf{MYOPIA} therefore provides a framework for diagnosing hierarchical visual classification and motivates testing tree designs that prioritize perceptual answerability and recovery from early errors. Code is available at https://github.com/islammesabah/VLM_Classification.

cs.CV↗

Puffin: Probabilistic Learning of Spatial Detail From Coarse Observations

High-resolution socioeconomic variables are important for applications such as urban planning, public health, disaster response, and resource allocation. In practice, however, these variables are often observed only at a coarse spatial resolution. We introduce Puffin, a probabilistic framework for statistical disaggregation that raises the resolution of coarse totals using high-resolution satellite embeddings as covariates. Instead of predicting a single value for each fine-resolution subregion, Puffin learns a probability distribution and is trained through an aggregation-aware likelihood. At inference, Puffin conditions these predictions on the observed regional total and splits it among the subregions. The resulting fine-scale estimates are consistent with the observed aggregate and come with calibrated uncertainty, without requiring fine-resolution labels for training. We evaluate Puffin on German and US census, employment, and election data across population, jobs, and other count variables, and study when statistical disaggregation succeeds or fails across regions, countries, and targets.

cs.LG↗

Accounting for Bias Enables Sustainable LLM Evaluation

LLM-as-a-judge has become the de facto standard for scalable, subjective evaluation, yet current leaderboards compensate for systematic measurement bias by running ever more comparisons, an approach that is both statistically unsound and computationally wasteful. The root cause is an incomplete measurement model, treating LLM judges as neutral, interchangeable instruments ignores documented biases like position bias, verbosity bias, judge severity, and self-enhancement, that no volume of additional data can eliminate. We propose a unified latent variable framework that jointly models pairwise and ordinal data while explicitly correcting for these confounders, recovering reliable rankings from substantially fewer comparisons. Because fitting this model costs negligible compute relative to a single round of LLM inference, bias correction is not only more statistically rigorous but also a more sustainable approach to trustworthy evaluation.

cs.AI↗

CLAM: Causal Spatial Disaggregation to Infer Local Effects From Coarse Data

Learning fine-grained spatial patterns from coarse-resolution data is challenging, especially in causal settings where high-resolution effects must be inferred from aggregated interventions and outcomes. We introduce CLAM, a method for estimating localized causal effects from coarse observations by exploiting high-resolution contextual covariates that modulate these effects. By jointly learning the causal mechanism and a disaggregation mapping, CLAM captures interactions that are missed when addressing these problems independently. The method supports localized effect estimation, counterfactual reasoning, and principled outcome disaggregation, and reliably captures spatially varying causal effects across diverse settings. This is particularly relevant for applications such as public health and environmental policy, where decisions are made at broad scales despite substantial local heterogeneity. Code is available at https://github.com/gerritgr/clam

cs.LG↗

Orca: Neural Operators for Causal Reasoning in Continuous Time

Structural causal models are the standard language for reasoning about interventions and counterfactuals, but they describe static variables, typically measured once, and usually forbid cyclic dependencies. Many systems we care about, such as patients, climates, and economies, instead evolve continuously in time, are observed at irregular time points, and contain feedback loops. We argue that neural operator learning provides a natural foundation for causal reasoning in this setting, and propose Orca, a framework in which each node of the causal graph is a function of time and each mechanism is a learned map between function spaces. We extend existing neural operator architectures to express causal mechanisms: a mechanism computes the function value of a node from its parent nodes by taking several parent functions as input, respects the arrow of time, and treats latent exogenous noise as a function that can be inferred and reused for counterfactuals. We formalize the model class and demonstrate counterfactual reasoning on synthetic continuous-time examples. Code is available at https://github.com/gerritgr/orca

cs.AI↗

CleanSurvival: Automated data preprocessing for time-to-event models using reinforcement learning

Data preprocessing is often paid little attention in machine learning, despite its potentially significant impact on model performance. While automated machine learning pipelines are starting to recognise and integrate data preprocessing into their solutions for classification and regression tasks, this integration is lacking for more specialised tasks like time-to-event models for censored data. As a result, survival analysis not only faces the general challenges of data preprocessing but also suffers from the lack of tailored, automated solutions in this area. To address this gap, this paper presents CleanSurvival, a reinforcement-learning-based solution for optimizing preprocessing pipelines, extended specifically for survival analysis. The framework can handle continuous and categorical variables. It builds upon Learn2Clean's Q-learning to select which combination of data imputation, outlier detection and feature extraction techniques achieves optimal performance for a Cox, random forest, neural network or user-supplied time-to-event model. Experimental benchmarks on real-world datasets show that the Q-learning-based data preprocessing can improve predictive performance relative to simple baselines, while runtime behaviour is condition-dependent and most clearly interpretable in the best-covered benchmark cells. Furthermore, a simulation study demonstrates effectiveness across different types and levels of missingness and noise. With an increase in the use of machine learning, it becomes important to generalise AutoML pipelines to a variety of models now present, including survival analysis. Tools like CleanSurvival, which integrate preprocessing for survival analysis, can make survival studies faster and easier to perform, while also yielding more robust results.

cs.LG↗

When Context Compensates for Sparse Event History: AlphaEarth for Spatio-Temporal Point-Process Forecasting

Spatio-temporal point-process models must often generalise across space when local event histories are sparse. We study whether exogenous spatial context can compensate in such regimes. Using a fixed log-Gaussian Cox process backbone, we compare an event-only model with the same model augmented by AlphaEarth embeddings as linear spatial context. We evaluate spatial transfer on emergency medical services (EMS) forecasting across eight held-out regions, fixed forecast anchors, and a sweep over history length $w$, using only AlphaEarth (AE) embeddings available strictly before each anchor. AE improves out-of-region predictive performance across all history regimes, with the largest gains under scarce histories: approximately $2$--$6\times$ multiplicative improvements at $1-2$ weeks, tapering to roughly $10$--$20\%$ at $w=20$--$104$ weeks. These results show that contextual information can substantially stabilise spatially transferred point-process forecasts when event history is limited.

cs.LG↗

HawkesNest: A Multi-Axis Synthetic Benchmark for Spatiotemporal Pattern Complexity

Evaluation of spatiotemporal point process (STPP) models relies heavily on opaque real-world datasets, where latent generative structure is unknown and model failures are difficult to attribute. We introduce HawkesNest, a generator-aligned benchmark for controlled spatiotemporal pattern complexity built on a multivariate Hawkes backbone. HawkesNest defines four complexity axes: space--time entanglement, background heterogeneity, cross-type interaction, and domain topology. Each axis is associated with a deterministic index computed from the latent data-generating mechanism. By varying these axes while holding global rate, stability, and simulation budget fixed, HawkesNest enables diagnostic stress tests of STPP models under known structural difficulty. We verify that the indices are monotone and nearly orthogonal under controlled sweeps. We illustrate its use by showing that Hawkes-family baselines degrade under joint heterogeneity--entanglement complexity, even though they are structurally aligned with the Hawkes data-generating backbone. We further show that HawkesNest exposes neural-model sensitivity: AutoSTPP remains vulnerable under isolated increases in space--time entanglement. Code. Available at https://github.com/YahyaAalaila/HawkesNest

cs.LG↗

BioDisco: Multi-agent hypothesis generation with dual-mode evidence, iterative feedback and temporal evaluation

Identifying novel hypotheses is essential to scientific research, yet this process risks being overwhelmed by the sheer volume and complexity of available information. Existing automated methods often struggle to generate novel and evidence-grounded hypotheses, lack robust iterative refinement and rarely undergo rigorous temporal evaluation for future discovery potential. To address this, we propose BioDisco, a multi-agent framework that draws upon language model-based reasoning and a dual-mode evidence system (biomedical knowledge graphs and automated literature retrieval) for grounded novelty, integrates an internal scoring and feedback loop for iterative refinement, and validates performance through pioneering temporal and human evaluations and a Bradley-Terry paired comparison model to provide statistically-grounded assessment. Our evaluations demonstrate superior novelty and significance over ablated configurations and generalist biomedical agents. Designed for flexibility and modularity, BioDisco allows seamless integration of custom language models or knowledge graphs, and can be run with just a few lines of code.

cs.AI↗

MEDAKA: Construction of Biomedical Knowledge Graphs Using Large Language Models

Knowledge graphs (KGs) are increasingly used to represent biomedical information in structured, interpretable formats. However, existing biomedical KGs often focus narrowly on molecular interactions or adverse events, overlooking the rich data found in drug leaflets. In this work, we present (1) a hackable, end-to-end pipeline to create KGs from unstructured online content using a web scraper and an LLM; and (2) a curated dataset, MEDAKA, generated by applying this method to publicly available drug leaflets. The dataset captures clinically relevant attributes such as side effects, warnings, contraindications, ingredients, dosage guidelines, storage instructions and physical characteristics. We evaluate it through manual inspection and with an LLM-as-a-Judge framework, and compare its coverage with existing biomedical KGs and databases. We expect MEDAKA to support tasks such as patient safety monitoring and drug recommendation. The pipeline can also be used for constructing KGs from unstructured texts in other domains. Code and dataset are available at https://github.com/medakakg/medaka.

cs.AI↗

The Power of Stories: Narrative Priming Shapes How LLM Agents Collaborate and Compete

According to Yuval Noah Harari, large-scale human cooperation is driven by shared narratives that encode common beliefs and values. This study explores whether such narratives can similarly nudge LLM agents toward collaboration. We use a finitely repeated public goods game in which LLM agents choose either cooperative or egoistic spending strategies. We prime agents with stories highlighting teamwork to different degrees and test how this influences negotiation outcomes. Our experiments explore four questions:(1) How do narratives influence negotiation behavior? (2) What differs when agents share the same story versus different ones? (3) What happens when the agent numbers grow? (4) Are agents resilient against self-serving negotiators? We find that story-based priming significantly affects negotiation strategies and success rates. Common stories improve collaboration, benefiting each agent. By contrast, priming agents with different stories reverses this effect, and those agents primed toward self-interest prevail. We hypothesize that these results carry implications for multi-agent system design and AI alignment.

cs.AI↗

When Counterfactual Reasoning Fails: Chaos and Real-World Complexity

Counterfactual reasoning, a cornerstone of human cognition and decision-making, is often seen as the 'holy grail' of causal learning, with applications ranging from interpreting machine learning models to promoting algorithmic fairness. While counterfactual reasoning has been extensively studied in contexts where the underlying causal model is well-defined, real-world causal modeling is often hindered by model and parameter uncertainty, observational noise, and chaotic behavior. The reliability of counterfactual analysis in such settings remains largely unexplored. In this work, we investigate the limitations of counterfactual reasoning within the framework of Structural Causal Models. Specifically, we empirically investigate \emph{counterfactual sequence estimation} and highlight cases where it becomes increasingly unreliable. We find that realistic assumptions, such as low degrees of model uncertainty or chaotic dynamics, can result in counterintuitive outcomes, including dramatic deviations between predicted and true counterfactual trajectories. This work urges caution when applying counterfactual reasoning in settings characterized by chaos and uncertainty. Furthermore, it raises the question of whether certain systems may pose fundamental limitations on the ability to answer counterfactual questions about their behavior.

cs.LG↗

Auto-encoding Molecules: Graph-Matching Capabilities Matter

Autoencoders are effective deep learning models that can function as generative models and learn latent representations for downstream tasks. The use of graph autoencoders - with both encoder and decoder implemented as message passing networks - is intriguing due to their ability to generate permutation-invariant graph representations. However, this approach faces difficulties because decoding a graph structure from a single vector is challenging, and comparing input and output graphs requires an effective permutation-invariant similarity measure. As a result, many studies rely on approximate methods. In this work, we explore the effect of graph matching precision on the training behavior and generation capabilities of a Variational Autoencoder (VAE). Our contribution is two-fold: (1) we propose a transformer-based message passing graph decoder as an alternative to a graph neural network decoder, that is more robust and expressive by leveraging global attention mechanisms. (2) We show that the precision of graph matching has significant impact on training behavior and is essential for effective de novo (molecular) graph generation. Code is available at https://github.com/mcunow/graph-matching

cs.LG↗