arXiv subjects

Konstantinos Voudouris

Publications and source records attributed to Konstantinos Voudouris.

At least 19 recordsLinked to original sources

arXiv2026

Can AI agents conduct open-ended AI research? Early evidence from two case studies

Forecasts of explosive AI progress hinge on AI agents automating AI research. But evidence on whether agents can carry out open-ended AI research is thin. Current evaluations either test agents on narrow, verifiable tasks, which excludes open-ended research, or submit AI-generated papers to blind peer review, which is overstretched, stochastic, and suffers from poor review quality. We introduce a third way to measure progress towards AI R\&D automation. An agent takes on the central, open-ended research question of a high-quality unpublished paper, and the paper's original authors grade its output. We call these shadow evaluations. We ran shadow evaluations on two unpublished NeurIPS 2026 submissions, giving frontier agents six days and thousands of dollars of compute. The agents completed all of the engineering without human help, yet could not make substantial progress towards answering the research questions. As a result, both papers were unambiguously rejected by the authors. We identify five recurring failure modes: poor judgment about the bar for publishable research, uncreative responses to shortcomings in the research design, ineffective backtracking from dead ends, poor resource awareness, and instruction drift. A robustness check with a second model and scaffold reproduced these failures. We release the expert reviews, survey responses, agent repositories, and logs. Our results provide early evidence that today's agents can do the engineering of AI research, but struggle with critical parts of the research lifecycle.

cs.AI↗

arXiv2026

Item Response Theory for AI Safety

Language models differ in how safely they behave and these differences are measured by safety benchmarks. But aggregated benchmark scores are hard to trust and interpret, because benchmarks duplicate one another, correlate heavily, and models may sandbag when they detect evaluation. To address these issues, we draw on Item Response Theory (IRT), a statistical toolkit for measuring these latents from performance on items with inferred psychometric properties. We fit IRT models to eight safety benchmarks across 192 language models, the largest psychometric analysis of LLM safety evaluations to date, and contribute three results. First, we find that three interpretable factors of refusal strictness, truthfulness, and contextual harm explain most of the variance between models across benchmarks. Second, psychometrically selected items recover full benchmark scores with lower error than random subsets of the same size, and roughly ten adaptively chosen items suffice for several individual benchmarks, cutting evaluation cost by 97-99%. Third, IRT supports audits of individual models, showing that it can be used to detect naive sandbagging and changes of model behind APIs. Overall, we show IRT is a ready-made toolkit for reading, reducing, and auditing safety benchmarks, which we recommend frontier labs and evaluators adopt.

cs.AI↗

arXiv2026

Persona Cartography: Charting Language Model Personality Traits in Weight Space

Large language models exhibit recurring behavioural patterns -- personas -- that shape generalisation and safety, but we lack reliable tools for decomposing, measuring, and controlling them. Our central insight is to treat personas as positions in a space of behavioural traits, using the OCEAN framework to describe model personas in terms of Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. We train low-rank adapters to amplify or suppress individual traits, and evaluate their effects using an LLM-judge calibrated against a human-validated panel, trait-specific multiple-choice benchmarks, and standard capability evaluations. Across six models from three families (4B-32B), we find that each adapter moves its target trait largely monotonically with scale, combines approximately additively with other adapters to construct mixed personas, and preserves performance on capability benchmarks at moderate scales. We further show that the induced trait axes affect safety-relevant behaviour in downstream evaluations: for example, moving along neuroticism and agreeableness axes affects frustration and sycophancy respectively. We also introduce an unsupervised psychometric pipeline that recovers four interpretable behavioural factors (tone, initiative, didacticism, epistemic caution) from model rollouts. Persona control can then be considered in terms of learning, scaling, and composing traits in weight space, providing a bridge between personality measurement, model editing, and safety.

cs.AI↗

arXiv2026

Can Vision Language Models Learn Intuitive Physics from Interaction?

Pre-trained vision language models do not have good intuitions about the physical world. Recent work has shown that supervised fine-tuning can improve model performance on simple physical tasks. However, fine-tuned models do not appear to learn robust physical rules that can generalize to new contexts. Based on research in cognitive science, we hypothesize that models need to interact with an environment to properly learn its physical dynamics. We train models that learn through interaction with a simulated environment using reinforcement learning. While learning from interaction allows models to improve their within-task performance, it fails to produce models with generalizable physical intuitions. We find that models trained on one task do not reliably generalize to related tasks, even if the tasks share visual statistics and physical principles, and regardless of whether the models are trained through interaction.

cs.LG↗

arXiv2026

Post-training makes large language models less human-like

Large language models (LLMs) are increasingly used as surrogates for human participants, but it remains unclear which models best capture human behavior and why. To address this, we introduce Psych-201, a novel dataset that enables us to measure behavioral alignment at scale. We find that post-training -- the stage that turns base models into useful assistants -- consistently reduces alignment with human behavior across model families, sizes, and objectives. Moreover, this misalignment widens in newer model generations even as base models continue to improve. Finally, we find that persona-induction -- a popular technique for eliciting human-like behavior by conditioning models on participant-specific information -- does not improve predictions at the level of individuals. Taken together, our results suggest that the very processes that are currently employed to turn LLMs into useful assistants also make them less accurate models of human behavior.

cs.CL↗

arXiv2026

In-Context Function Learning in Large Language Models

Large language models (LLMs) can learn from a few demonstrations provided at inference time. We study this in-context learning phenomenon through the lens of Gaussian Processes (GPs). We build controlled experiments where models observe sequences of multivariate scalar-valued function samples drawn from known GP priors. We evaluate prediction error in relation to the number of demonstrations and compare against two principled references: (i) an empirical GP-regression learner that gives a lower bound on achievable error, and (ii) the expected error of a 1-nearest-neighbor (1-NN) rule, which gives a data-driven upper bound. Across model sizes, we find that LLM learning curves are strongly influenced by the function-generating kernels and approach the GP lower bound as the number of demonstrations increases. We then study the inductive biases of these models using a likelihood-based analysis. We find that LLM predictions are most likely under less smooth GP kernels. Finally, we explore whether post-training can shift these inductive biases and improve sample-efficiency on functions sampled from GPs with smoother kernels. We find that both reinforcement learning and supervised fine-tuning can effectively shift inductive biases in the direction of the training data. Together, our framework quantifies the extent to which LLMs behave like GP learners and provides tools for steering their inductive biases for continuous function learning tasks.

cs.LG↗

arXiv2026

Measuring What AI Systems Might Do: Towards A Measurement Science in AI

Scientists, policy-makers, business leaders, and members of the public care about what modern artificial intelligence systems are disposed to do. Yet terms such as capabilities, propensities, skills, values, and abilities are routinely used interchangeably and conflated with observable performance, with AI evaluation practices rarely specifying what quantity they purport to measure. We argue that capabilities and propensities are dispositional properties - stable features of systems characterised by counterfactual relationships between contextual conditions and behavioural outputs. Measuring a disposition requires (i) hypothesising which contextual properties are causally relevant, (ii) independently operationalising and measuring those properties, and (iii) empirically mapping how variation in those properties affects the probability of the behaviour. Dominant approaches to AI evaluation, from benchmark averages to data-driven latent-variable models such as Item Response Theory, bypass these steps entirely. Building on ideas from philosophy of science, measurement theory, and cognitive science, we develop a principled account of AI capabilities and propensities as dispositions, show why prevailing evaluation practices fail to measure them, and outline what disposition-respecting, scientifically defensible AI evaluation would require.

cs.CY↗

arXiv2025

Inferring Capabilities from Task Performance with Bayesian Triangulation

As machine learning models become more general, we need to characterise them in richer, more meaningful ways. We describe a method to infer the cognitive profile of a system from diverse experimental data. To do so, we introduce measurement layouts that model how task-instance features interact with system capabilities to affect performance. These features must be triangulated in complex ways to be able to infer capabilities from non-populational data -- a challenge for traditional psychometric and inferential tools. Using the Bayesian probabilistic programming library PyMC, we infer different cognitive profiles for agents in two scenarios: 68 actual contestants in the AnimalAI Olympics and 30 synthetic agents for O-PIAAGETS, an object permanence battery. We showcase the potential for capability-oriented evaluation.

cs.AI↗

arXiv2025

Exploring Major Transitions in the Evolution of Biological Cognition With Artificial Neural Networks

Transitional accounts of evolution emphasise a few changes that shape what is evolvable, with dramatic consequences for derived lineages. More recently it has been proposed that cognition might also have evolved via a series of major transitions that manipulate the structure of biological neural networks, fundamentally changing the flow of information. We used idealised models of information flow, artificial neural networks (ANNs), to evaluate whether changes in information flow in a network can yield a transitional change in cognitive performance. We compared networks with feed-forward, recurrent and laminated topologies, and tested their performance learning artificial grammars that differed in complexity, controlling for network size and resources. We documented a qualitative expansion in the types of input that recurrent networks can process compared to feed-forward networks, and a related qualitative increase in performance for learning the most complex grammars. We also noted how the difficulty in training recurrent networks poses a form of transition barrier and contingent irreversibility -- other key features of evolutionary transitions. Not all changes in network topology confer a performance advantage in this task set. Laminated networks did not outperform non-laminated networks in grammar learning. Overall, our findings show how some changes in information flow can yield transitions in cognitive performance.

cs.AI↗

arXiv2025

PredictaBoard: Benchmarking LLM Score Predictability

Despite possessing impressive skills, Large Language Models (LLMs) often fail unpredictably, demonstrating inconsistent success in even basic common sense reasoning tasks. This unpredictability poses a significant challenge to ensuring their safe deployment, as identifying and operating within a reliable "safe zone" is essential for mitigating risks. To address this, we present PredictaBoard, a novel collaborative benchmarking framework designed to evaluate the ability of score predictors (referred to as assessors) to anticipate LLM errors on specific task instances (i.e., prompts) from existing datasets. PredictaBoard evaluates pairs of LLMs and assessors by considering the rejection rate at different tolerance errors. As such, PredictaBoard stimulates research into developing better assessors and making LLMs more predictable, not only with a higher average performance. We conduct illustrative experiments using baseline assessors and state-of-the-art LLMs. PredictaBoard highlights the critical need to evaluate predictability alongside performance, paving the way for safer AI systems where errors are not only minimised but also anticipated and effectively mitigated. Code for our benchmark can be found at https://github.com/Kinds-of-Intelligence-CFI/PredictaBoard

cs.CL↗

arXiv2025

Testing the Limits of Fine-Tuning for Improving Visual Cognition in Vision Language Models

Pre-trained vision language models still fall short of human visual cognition. In an effort to improve visual cognition and align models with human behavior, we introduce visual stimuli and human judgments on visual cognition tasks, allowing us to systematically evaluate performance across cognitive domains under a consistent environment. We fine-tune models on ground truth data for intuitive physics and causal reasoning and find that this improves model performance in the respective fine-tuning domain. Furthermore, it can improve model alignment with human behavior. However, we find that task-specific fine-tuning does not contribute to robust human-like generalization to data with other visual characteristics or to tasks in other cognitive domains.

cs.LG↗

arXiv2025

Centaur: a foundation model of human cognition

Establishing a unified theory of cognition has been a major goal of psychology. While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in its entirety. A first step in this direction is to create a model that can predict human behavior in a wide range of settings. Here we introduce Centaur, a computational model that can predict and simulate human behavior in any experiment expressible in natural language. We derived Centaur by finetuning a state-of-the-art language model on a novel, large-scale data set called Psych-101. Psych-101 reaches an unprecedented scale, covering trial-by-trial data from over 60,000 participants performing over 10,000,000 choices in 160 experiments. Centaur not only captures the behavior of held-out participants better than existing cognitive models, but also generalizes to new cover stories, structural task modifications, and entirely new domains. Furthermore, we find that the model's internal representations become more aligned with human neural activity after finetuning. Taken together, our results demonstrate that it is possible to discover computational models that capture human behavior across a wide range of domains. We believe that such models provide tremendous potential for guiding the development of cognitive theories and present a case study to demonstrate this.

cs.LG↗

arXiv2025

Cognitive Science-Inspired Evaluation of Core Capabilities for Object Understanding in AI

One of the core components of our world models is 'intuitive physics' - an understanding of objects, space, and causality. This capability enables us to predict events, plan action and navigate environments, all of which rely on a composite sense of objecthood. Despite its importance, there is no single, unified account of objecthood, though multiple theoretical frameworks provide insights. In the first part of this paper, we present a comprehensive overview of the main theoretical frameworks in objecthood research - Gestalt psychology, enactive cognition, and developmental psychology - and identify the core capabilities each framework attributes to object understanding, as well as what functional roles they play in shaping world models in biological agents. Given the foundational role of objecthood in world modelling, understanding objecthood is also essential in AI. In the second part of the paper, we evaluate how current AI paradigms approach and test objecthood capabilities compared to those in cognitive science. We define an AI paradigm as a combination of how objecthood is conceptualised, the methods used for studying objecthood, the data utilised, and the evaluation techniques. We find that, whilst benchmarks can detect that AI systems model isolated aspects of objecthood, the benchmarks cannot detect when AI systems lack functional integration across these capabilities, not solving the objecthood challenge fully. Finally, we explore novel evaluation approaches that align with the integrated vision of objecthood outlined in this paper. These methods are promising candidates for advancing from isolated object capabilities toward general-purpose AI with genuine object understanding in real-world contexts.

cs.AI↗

arXiv2025

Bringing Comparative Cognition To Computers

Researchers are increasingly subjecting artificial intelligence systems to psychological testing. But to rigorously compare their cognitive capacities with humans and other animals, we must avoid both over- and under-stating our similarities and differences. By embracing a comparative approach, we can integrate AI cognition research into the broader cognitive sciences.

cs.AI↗

arXiv2025

metabench -- A Sparse Benchmark of Reasoning and Knowledge in Large Language Models

Large Language Models (LLMs) vary in their abilities on a range of tasks. Initiatives such as the Open LLM Leaderboard aim to quantify these differences with several large benchmarks (sets of test items to which an LLM can respond either correctly or incorrectly). However, high correlations within and between benchmark scores suggest that (1) there exists a small set of common underlying abilities that these benchmarks measure, and (2) items tap into redundant information and the benchmarks may thus be considerably compressed. We use data from n > 5000 LLMs to identify the most informative items of six benchmarks, ARC, GSM8K, HellaSwag, MMLU, TruthfulQA and WinoGrande (with d = 28,632 items in total). From them we distill a sparse benchmark, metabench, that has less than 3% of the original size of all six benchmarks combined. This new sparse benchmark goes beyond point scores by yielding estimators of the underlying benchmark-specific abilities. We show that these estimators (1) can be used to reconstruct each original individual benchmark score with, on average, 1.24% root mean square error (RMSE), (2) reconstruct the original total score with 0.58% RMSE, and (3) have a single underlying common factor whose Spearman correlation with the total score is r = 0.94.

cs.CL↗

arXiv2025

The Animal-AI Environment: A Virtual Laboratory For Comparative Cognition and Artificial Intelligence Research

The Animal-AI Environment is a unique game-based research platform designed to facilitate collaboration between the artificial intelligence and comparative cognition research communities. In this paper, we present the latest version of the Animal-AI Environment, outlining several major features that make the game more engaging for humans and more complex for AI systems. These features include interactive buttons, reward dispensers, and player notifications, as well as an overhaul of the environment's graphics and processing for significant improvements in agent training time and quality of the human player experience. We provide detailed guidance on how to build computational and behavioural experiments with the Animal-AI Environment. We present results from a series of agents, including the state-of-the-art deep reinforcement learning agent Dreamer-v3, on newly designed tests and the Animal-AI Testbed of 900 tasks inspired by research in the field of comparative cognition. The Animal-AI Environment offers a new approach for modelling cognition in humans and non-human animals, and for building biologically inspired artificial intelligence.

cs.AI↗

arXiv2025

Predictable Artificial Intelligence

We introduce the fundamental ideas and challenges of Predictable AI, a nascent research area that explores the ways in which we can anticipate key validity indicators (e.g., performance, safety) of present and future AI ecosystems. We argue that achieving predictability is crucial for fostering trust, liability, control, alignment and safety of AI ecosystems, and thus should be prioritised over performance. We formally characterise predictability, explore its most relevant components, illustrate what can be predicted, describe alternative candidates for predictors, as well as the trade-offs between maximising validity and predictability. To illustrate these concepts, we bring an array of illustrative examples covering diverse ecosystem configurations. Predictable AI is related to other areas of technical and non-technical AI research, but have distinctive questions, hypotheses, techniques and challenges. This paper aims to elucidate them, calls for identifying paths towards a landscape of predictably valid AI systems and outlines the potential impact of this emergent field.

cs.AI↗

arXiv2025

A little less conversation, a little more action, please: Investigating the physical common-sense of LLMs in a 3D embodied environment

As general-purpose tools, Large Language Models (LLMs) must often reason about everyday physical environments. In a question-and-answer capacity, understanding the interactions of physical objects may be necessary to give appropriate responses. Moreover, LLMs are increasingly used as reasoning engines in agentic systems, designing and controlling their action sequences. The vast majority of research has tackled this issue using static benchmarks, comprised of text or image-based questions about the physical world. However, these benchmarks do not capture the complexity and nuance of real-life physical processes. Here we advocate for a second, relatively unexplored, approach: 'embodying' the LLMs by granting them control of an agent within a 3D environment. We present the first embodied and cognitively meaningful evaluation of physical common-sense reasoning in LLMs. Our framework allows direct comparison of LLMs with other embodied agents, such as those based on Deep Reinforcement Learning, and human and non-human animals. We employ the Animal-AI (AAI) environment, a simulated 3D virtual laboratory, to study physical common-sense reasoning in LLMs. For this, we use the AAI Testbed, a suite of experiments that replicate laboratory studies with non-human animals, to study physical reasoning capabilities including distance estimation, tracking out-of-sight objects, and tool use. We demonstrate that state-of-the-art multi-modal models with no finetuning can complete this style of task, allowing meaningful comparison to the entrants of the 2019 Animal-AI Olympics competition and to human children. Our results show that LLMs are currently outperformed by human children on these tasks. We argue that this approach allows the study of physical reasoning using ecologically valid experiments drawn directly from cognitive science, improving the predictability and reliability of LLMs.

cs.AI↗