AI NEWS JOURNAL

AI, IN PERSPECTIVE

Search AI news

Search headlines, summaries and articles to connect the dots.

2026-11-13

arXiv cs.CL13:00

An LLM Is All You Need: Data-Efficient Adaptation to Full-Duplex, Word-Level Spoken Dialogue

arXiv:2611.14774v1 Announce Type: new Abstract: Full-duplex spoken dialogue, in which a system listens while it speaks and can backchannel, yield the floor, and be interrupted, is usually obtained by training on hundreds of thousands to millions of hours of speech. This is costly, rewrites the language model that supplies the system's knowledge, and still leaves turn-taking either implicit or decided coarsely rather than word by word. We show that a frozen, general-purpose multimodal LLM can instead be adapted into a word-level full-duplex system with little data. Our system is a thinker-talker cascade on an 80 ms frame clock. The thinker keeps the LLM frozen and trains only lightweight adapters and a few new token embeddings (97M parameters); at every frame it emits one token of agent text, with pauses, yields, and onsets represented as tokens of the same vocabulary and kept legal by a finite-state machine. A four-stage curriculum, each stage anchored to the model's own behavior under the previous format, performs the conversion in about 30 GPU-hours. The talker reads only the thinker's discrete tokens, so the two halves are trained independently. On Full-Duplex-Bench and InteractionBench, the system is competitive with open-source full-duplex systems built through large-scale speech training: among them, it is the least likely to take the floor when the user merely pauses, backchannels most often, and leads in spoken-task accuracy, groundedness, and coherence over long conversations.

arXiv cs.DC13:00

Autonomous Mobile Robots on a Computing Field

arXiv:2611.16652v1 Announce Type: new Abstract: We introduce CABOT, a computational model for autonomous mobile robots in which the discrete plane the robots operate on is not a passive stage but an active computing medium. In the classical Look-Compute-Move (LCM) model every increase in computational capability is obtained by strengthening the robots themselves, with lights, memory or wider visibility, at a cost paid once per robot. CABOT strengthens the environment instead: each cell of the grid holds a state from a finite set Q, the states evolve synchronously under a uniform local rule that also senses whether a robot stands on the cell, that is, the floor is a cellular automaton, and the Look operation returns the states of the visible cells in addition to the positions of the visible robots. The robots are as weak as we can make them: oblivious, silent, lightless, myopic with visibility range 1, without multiplicity detection and unable to write to the floor. The number of field states |Q|, rather than any robot capability, becomes the measure of the resources required. On an inert floor (|Q| = 1) such robots can neither form a line, nor gather from scattered positions, nor explore a grid. We show that a handful of states changes this. Line Formation from connected configurations is solvable with four states, and field-silently, so that the floor also reaches a fixed point, with five. Gathering from arbitrary, possibly disconnected, configurations is solvable with three states and Line Formation from arbitrary configurations with six; we prove that neither can be solved field-silently with any number of states. A single robot explores every finite grid forever with four states and every torus with two, which is optimal. Further rule sets, validated by simulation only, equalize the spacing of robots on a row, fill a pattern painted on the floor and guide robots through a maze, and indicate the range of the model.

arXiv cs.AI13:00

Emergent Vision-Shaped Thinking in Text-Only Reasoning through Multimodal Training

arXiv:2611.14942v1 Announce Type: new Abstract: Does multimodal training shape how a language model thinks, even when the input is text only? We test this with targeted causal ablations across 10 multimodal configurations from 6 model families, with matched text-only backbones and cross-family transfer controls. We contrast spatial-imagery tasks that invite mental scene construction with symbol-manipulation tasks solvable by symbolic rules alone. We identify a vision-language axis in hidden-state space and ablate its projection at inference. In Chameleon-7B, this ablation drops accuracy on BIG-Bench Hard (BBH) navigate by 27.6 pp while leaving boolean expressions intact (+5.6 pp), a 33.2 pp double dissociation. The effect is direction-specific and layer-localized, extends to mental-simulation tasks such as mental rotation, paper folding, and perspective taking, and is not explained by task difficulty, reasoning depth, or sequential update load. Together, these results reveal emergent vision-shaped thinking (EVT): text-only reasoning that causally recruits representations introduced or strengthened by vision-language training, without assuming an image-like internal format. The measured effect is spatially concentrated at 7-8B and broadens to non-spatial visual attribute binding in the tested 32B configuration. In dual-path unified models (BAGEL, Janus-Pro), the causal handle localizes to the generation pathway as a compact one-dimensional carrier. The understanding pathway also encodes spatial information in a higher-rank, probe-decodable form, but is not the causal handle. To separate decodability from causal use, we analyze concentration, alignment, and semantic location. These diagnostics explain why behavior, probing, and intervention can diverge. Multimodal training thus reshapes the geometry of text-only reasoning, not just how visual input is processed.

arXiv cs.LG13:00

Graph Label Alignment: A Diagnostic Atlas for Graph Classification

arXiv:2611.14760v1 Announce Type: new Abstract: Graph classification benchmarks vary widely in difficulty: some are solved by simple pooled baselines, while others remain challenging for GNNs and graph transformers. We take a data-centric view and ask whether benchmark success is driven by genuine structure--feature reasoning or by label signal already accessible from restricted marginal views. We introduce graph label alignment, a lightweight diagnostic that measures label accessibility through two supervised probes computed without training a graph encoder: structural label alignment (SLA), based on permutation-invariant topological and spectral summaries, and feature label alignment (FLA), based on pooled node features with edges removed. Across real graph classification benchmarks and multiple pooled backbones, the resulting SLA--FLA plane strongly organizes test accuracy, with low-alignment datasets forming a consistent regime where standard pooled pipelines struggle. To probe what lies beyond marginal accessibility, we introduce controlled interaction-dominated stress tests, including heterophilic ego-graphs and synthetic Twins tasks whose structural and feature marginals are matched across classes but whose role--feature couplings differ. These results show that high alignment explains much of easy benchmark performance, while low alignment reveals regimes where success requires preserving structure--feature correspondence beyond standard pooled representations.

arXiv cs.LG13:00

KernelMosaic: An Open Dataset and Agentic Framework for CUDA Kernel Generation

arXiv:2611.14604v1 Announce Type: new Abstract: Training large language models to generate efficient CUDA kernels requires supervision spanning diverse computational requirements and iterative code optimization. Constructing this supervision is challenging because of heterogeneous specifications, and useful implementations emerge through iterative complicated testing and refinement. We introduce KernelMosaic, an open dataset pairing diverse CUDA tasks with verified solutions and trajectories. Human-guided review and diversity-aware selection curate tasks expressed as natural-language descriptions, reference programs, mathematical specifications, and their intersections. To build the solutions and trajectories, our agentic framework guides a strong LLM through implementation, correctness repair, and performance optimization with reusable CUDA guidance and execution feedback. Supervised fine-tuning of a small open-source model on KernelMosaic raises correctness from 4%, 0%, and 0% to 88%, 82%, and 16% on KernelBench L1, L2, and L3, respectively, while increasing the fraction of tasks with kernels faster than PyTorch eager at every level. Further training experiments show that supervision from the curated trajectories substantially outperforms direct responses and promotes sustained testing and revision. These results demonstrate that KernelMosaic enables a general-purpose model to achieve substantial gains in CUDA kernel generation.

arXiv cs.CV13:00

Lines Are Ideas: Visual Creativity Assessment in Hyperbolic Space

arXiv:2611.14907v1 Announce Type: new Abstract: Assessing human creativity from drawings provides a scalable way to evaluate human creative ability. However, existing approaches either rely on costly expert annotation or treat creativity as a single regression target, overlooking its underlying structure. Psychological theories have long emphasized that creativity emerges from balancing novelty and appropriateness. Motivated by this principle, we recast creativity as structured semantic deviation between paired drawings, where novelty measures the extent of departure from the original base sketch, and appropriateness determines whether such departures remain meaningful. This formulation naturally leads to a hyperbolic representation, in which novelty is quantified by geodesic distance and appropriateness is evaluated conditionally via a learned decision boundary, further regulated by an abstraction-conditioned adaptive constraint. Experiments on AuDrA-Drawings demonstrate state-of-the-art alignment with expert judgments and strong robustness under rater and task shifts. Beyond performance gains, our framework reveals an interpretable radial organization of creative variation, offering a principled geometric foundation for automatic creativity assessment.

arXiv cs.CL13:00

LLMs and the Homogenization of Teaching

arXiv:2611.14828v1 Announce Type: new Abstract: Large language models (LLMs) are known to lack diversity in open-ended text generation, a limitation that is especially consequential in education, where homogeneity can narrow the kinds of thinking and disciplinary work students are invited to do. We study diversity in two educational generation tasks: math items and essay feedback. We compare frontier and open-weight models with human-authored reference distributions using lexical, semantic, and domain-grounded measures. For math, we examine the mathematical activities students are asked to perform; for feedback, we examine the revision work requested and the rhetorical strategies used to communicate it. We evaluate direct prompting, two diversity prompting methods, and post-training. Across both tasks, directly prompted model outputs are markedly less diverse than human-authored content--78% higher on lexical (Self-BLEU) and 51% higher on semantic (embedding) similarity--and represent a narrower range of instructional forms. Diversity prompting and post-training increase variation largely through surface features without broadening the instructional repertoire. Models inherit and amplify the most normative routines of human instruction. Overcoming mode collapse in education will require richer instructional data and investment in teachers' creative capacity and professional judgment.

arXiv cs.AI13:00

Provably Optimal HBT Legalization in Seconds: Certified Optima Reveal Where the Headroom Really Lies

arXiv:2611.14928v1 Announce Type: new Abstract: Hybrid bonding terminal (HBT) legalization assigns every terminal of a face-to-face 3D IC to its own pitch-grid site with minimum total Manhattan displacement. Open3DBench reports that a reinforcement-learning (RL) legalizer reduces displacement by 19-22% relative to a sliding-window (SW) heuristic, but none of its legalizers is publicly released. We ask how much displacement any legalizer can still recover. We build an open benchmark of 21 design/pitch instances from Open3DBench placement data, with reconstructed greedy, SW (ViaLegal-style) and BTAssign-style baselines, and GEO-Exact, an exact solver that runs successive shortest augmenting paths over lazily generated spatial candidate edges. An independent dual certificate proves its assignment optimal on all 21 instances (relative primal-dual gap at most $8.2 \times 10^{-16}$). Against these optima, our SW reconstruction with ViaLegal's default configuration is on average 0.011% above optimal (worst 0.119%), while on the 16 contended instances BTAssign-style, a density-guided heuristic (DGHL) and greedy are 3.0%, 6.9% and 64.5% above. Headroom does appear outside this regime: with 2x2 windows SW reaches 34.1%, and above about 58% utilization, on three of six swept designs, it reaches 21.1% on bp, 27.5% on swerv and 8.9% on jpeg (217 further certified instances, and ten bp points checked against an independent exact solver). If the published 19-22% gain were measured on instances like ours, SW would have to be at least 24-28% above optimal on average, far more than any SW variant we tested at nominal pitch; the RL paper's own comparison with the Hungarian optimum puts its SW 3.05% above optimal on average. GEO-Exact is also about an order of magnitude faster than DGHL and SW in our harness (geometric means 8.1x and 40x; a compiled core against Python baselines on a shared server). Code, optimal values and certificates are released.

arXiv cs.AI13:00

Termination Lag: Coding Agents Solve Early but Stop Late

arXiv:2611.14960v1 Announce Type: new Abstract: Coding agents can solve a task well before they stop working on it. We characterize \emph{termination lag}--the number of steps between an agent's first correct patch and the end of its run--across five models, two agent frameworks, and three SWE-bench datasets by recording and testing every distinct repository patch generated during execution. Our analysis reveals that 38--82% of successful runs achieve a correct solution on their initial source-code edit. However, agents continue to expend substantial effort after correct patches, with post-solution steps accounting for a median of 35--37% for Qwen3-Coder and 54--60% for Laguna-XS.2 in successful runs on SWE-bench Verified and Multilingual. Agents spend most of these additional steps verifying their work and exploring the repository. In a small fraction of cases (1--8%), this continued activity leads agents to undo their earlier correct patch, leaving the final solution incorrect.

arXiv cs.CL13:00

ToolTurns: Benchmarking LLM Agents in Multi-Turn Dialogue

arXiv:2611.14925v1 Announce Type: new Abstract: Tool-using language models are increasingly deployed in conversational settings, yet most large-scale tool-use benchmarks focus on isolated requests. We introduce ToolTurns, a benchmark of 2,620 problems spanning 1,827 tools and 11,004 API endpoints across 47 categories. Each dialogue begins with an intentionally underspecified request and continues through two context-dependent follow-ups, testing clarification, tool use, and retention of conversational context. We also introduce an open-source user simulator that reveals withheld context through dialogue and generates follow-up requests, with a checker--fixer pipeline for improving its turns. Across nine language models, the best agent resolves 69.6% of individual requests but completes only 39.8% of full dialogues. Under this evaluation protocol, every evaluated model receives lower success scores on follow-up requests than on the initial request, and dialogue-quality rubric scores and task success can rank models very differently. Paired turn-level evaluation shows that the checker--fixer reduces detected dialogue defects, and blind human evaluation favors the revised turns. ToolTurns provides a testbed for evaluating whether agents sustain task success across multi-turn dialogue with tools.

arXiv cs.CL13:00

What Do Tokenizer Comparisons Measure? Disentangling Granularity, Root Assignment, and Word-Level Readout across Amharic and Hausa

arXiv:2611.14818v1 Announce Type: new Abstract: Morphology-aware tokenizers are expected to help morphologically rich languages, and their reported gains over byte-pair encoding (BPE) are usually credited to their linguistic units. A comparison with BPE cannot support this attribution, since such a tokenizer also segments text more finely (granularity) and, on token-level tasks, changes how much the sub-token readout matters. We separate both factors from the effect of grouping words by their roots (root assignment) in Amharic, a Semitic language whose Ge'ez abugida hides consonantal roots inside syllabic glyphs, with Latin-script Hausa as an orthographic contrast. Our design combines a lossless root-and-pattern tokenizer with controls that change one factor at a time, in two independent implementations. In both, real root assignment improves Amharic named-entity recognition over scrambled roots (+4.9 and +4.7 F1), and root-frequency structure explains little of this gain. In small from-scratch models, the readout choice moves F1 by up to 13.4 points and reverses the sign of every comparison of the root-and-pattern arms with BPE, whereas in two pretrained encoders it moves F1 by less than one point. The granularity effect changes sign across languages and implementations. We recommend decomposing tokenizer effects, and reporting the readout, before a gain is attributed to linguistic structure.

arXiv cs.AI13:00

When Is Proof-Step Verification Worth Its Cost? Operating Requirements of Certify-or-Decline Verification for Open-Weight Mathematical Reasoning

arXiv:2611.14794v1 Announce Type: new Abstract: Proof-step verification rewrites a solver's answer into typed steps and uses LLM judges to certify or decline. We evaluate this pipeline on open-weight models across GPQA, AIME-2025, and a MATH-500 subset. Certification raises answer-key-match precision by 31 percentage points on the strong GPQA configuration and 46 on AIME, while answering only 35% and 21% of attempts. On near-ceiling MATH-500, gains are at most 5 points at a loss of 13-27 points of coverage. An output-token-matched self-consistency selector obtains 82.9% precision at 35% coverage, compared with the verifier's pooled 87.6% (paired difference: +4.8 points, 95% interval [-8.2, 18.5]). A diagnostic sample reveals a distinct failure mode: 52 of 88 completed problems end in invalid proof construction without ever reaching a judge. Finally, external-LLM acceptance of rendered certified chains is 61.0-67.6%, compared with 87.6% answer-key match. These audits lack human or formal-kernel ground truth. Open-weight auditor pairs show near-zero chance-corrected agreement, which measures consistency rather than accuracy against proof truth. These results motivate separate reporting of execution failures, selected-answer precision, and independently assessed proof validity. We provide per-configuration results and evaluation details in the supplement.

2026-11-11

arXiv cs.CL13:00

BioMem: Agentic Memory for Biomedical LLM Agents

arXiv:2611.13426v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated considerable potential in biomedical applications, while agentic memory systems have emerged as a promising approach to preserving and utilizing information across tasks and interactions. However, a fundamental question remains: how effectively and reliably do these systems retain, retrieve, update, and utilize biomedical information? In this study, we systematically investigate agentic memory in the biomedical domain, focusing on task effectiveness, retrieval quality, information fidelity, and long-term robustness. We evaluate five memory approaches across four LLMs in three representative settings: biomedical knowledge reasoning, clinical prediction, and long-term conversational memory. Our analyses reveal how memory construction, evidence relevance, and conversational history shape downstream biomedical performance, highlighting the distinct challenges of retrieving relevant information, preserving its fidelity, and effectively utilizing it for reasoning. Building on these findings, we introduce Hybrid Memory Retrieval and Utilization (HMRU), a training-free framework that jointly enhances evidence accessibility and task-aware utilization to improve biomedical reasoning and long-term conversational memory. HMRU achieves an average relative improvement of 8.6% across biomedical question-answering and conversational memory benchmarks, with additional gains in selected clinical prediction tasks. It further reduces conversational answer errors by 24.9% and closes 48.0% of the performance gap to full-context inference. Our findings provide a systematic characterization of biomedical agentic memory and practical insights into developing more effective and reliable memory-augmented LLM systems.

arXiv cs.CV13:00

Evidence-Grounded Iterative Reasoning for the Diagnosis of Diffuse Cystic Lung Diseases

arXiv:2611.12981v1 Announce Type: new Abstract: Overlapping imaging manifestations complicate the differential diagnosis of Diffuse Cystic Lung Diseases (DCLDs), while limited annotated data constrain supervised model development. Multimodal Large Language Models (MLLMs) can integrate imaging and clinical information, but their diagnostic reasoning remains limited by incomplete volumetric perception and insufficient grounding in disease-specific evidence. We propose DCLDs-EIR, an evidence-grounded iterative reasoning framework coupling diagnostic hypothesis refinement with multimodal evidence acquisition and verification. Multiview renderings of predicted lung and cyst segmentations characterize global cyst distributions and initialize the differential diagnosis. Competing hypotheses guide axial CT inspection to identify local findings that, together with clinical context, inform retrieval from a disease-specific knowledge graph. Accumulated patient observations and retrieved knowledge are jointly assessed for diagnostic support and refutation to revise the diagnostic state. Updated hypotheses and unresolved distinctions guide subsequent inspection, forming a feedback loop between evidence acquisition and hypothesis verification. Multicenter evaluation demonstrates that DCLDs-EIR consistently improves diagnostic accuracy over baselines and enables traceable differential diagnosis through explicit links between diagnostic hypotheses, patient observations, and disease-specific knowledge.

arXiv cs.LG13:00

Multimodal Graph Learning: A Survey of Models, Benchmarks, and Frontiers

arXiv:2611.12990v1 Announce Type: new Abstract: Multimodal attributed graphs (MAGs), in which nodes carry text, images, and other modalities while edges encode relations among entities, underlie e-commerce, social media, content platforms, and creative applications. Learning on MAGs has developed along two complementary lines: graph-centric methods redesign how modalities are fused, propagated, and connected within message passing, while foundation-model-centric methods pretrain transferable multimodal graph encoders or bring graph context into multimodal large language models for prediction and generation. In parallel, several benchmarks have appeared with different datasets and evaluation protocols. This survey takes an algorithm-centric and evaluation-grounded view of the field and makes three contributions. Modeling: we introduce a unified message-passing formulation and organize methods into five integration levels (feature, propagation, topology, learning, and foundation model), covering discriminative and graph-conditioned generative models; within this formulation we derive representative methods as special cases, compare their assumptions, expressiveness, and cost, and identify cross-modal sensitivity as a property that separates otherwise similar architectures. Datasets and evaluation: we review recent benchmarks, summarize their datasets and metrics, define model-independent descriptors (label homophily, average degree, clustering, and edge provenance), and map each evaluation protocol to the baselines it uses. Frontiers: we connect multimodal graph learning to AI agents, covering the multimodal graphs that agents perceive and remember and graph engineering, the practice of organizing agent workflows as explicit graphs; we then discuss open problems, concrete efforts for the community, and research directions, including when graph structure helps in the presence of strong multimodal foundation models.

arXiv cs.LG13:00

One Table, Many Queries: Benchmarking Generalist Tabular Models Across Generation, Imputation, and Prediction

arXiv:2611.13001v1 Announce Type: new Abstract: Generation, imputation, and prediction query different aspects of the same tabular distribution, but are usually evaluated separately. This leaves unclear which capabilities they share, how their training objectives interact, and whether one model can perform all three well. We introduce the One Table Benchmark (\ours{}), spanning 56 datasets, 46 registered methods, and 20 capability axes under aligned data splits and explicit task contracts. We compare task ranks on aligned support, train five backbones under all seven non-empty objective combinations on five datasets with three seeds, and contrast fixed-model profiles with specialist portfolios. Rank comparisons connect imputation and prediction through numerical recovery and dependence preservation, and link generation to both through predictive utility; distributional fidelity follows different, sometimes opposing, rankings. Objective interventions show the largest pooled gains from including the recipient's own objective. Generation has the broadest coverage of positive pooled auxiliary effects, while prediction shows potential to improve imputation and synthetic-data utility, with responses that vary across backbones. Cross-addition effects reveal that objective benefits need not add and depend on the objectives already present. Fixed-model comparisons expose corresponding utility--fidelity trade-offs: some generalists surpass specialist portfolios on utility, while fidelity continues to favor specialization. Frozen foundation models show strong prediction and numerical point imputation, together with promising synthetic-data utility, highlighting their potential across all three tasks. These findings motivate unified tabular models that share useful conditional information while improving categorical reconstruction and faithful joint sampling.

arXiv cs.CL13:00

SELD: Detecting Silent Evidence Loss in Context Compression beyond the Answer

arXiv:2611.12903v1 Announce Type: new Abstract: Context compression is now routine for language models with long inputs, but a compressed context is useful only if it keeps the evidence for the answer. Existing checks compare the reader model's answer or its most likely words, and thus miss a frequent failure, silent evidence loss: key evidence is removed while the answer stays the same, leaving the answer unsupported and risky for other models that reuse the context. To address this challenge, we propose SELD (Silent Evidence Loss Detector), which needs no training and no labels. Its principle is that when evidence is lost, a reader model's preferences among the answers it leaves unsaid shift even if its answer does not, so comparing these preferences before and after compression reveals losses that answer checks cannot see. We also build SEL-Pairs, a benchmark of length-matched evidence and control deletions from three question-answering datasets. Quantitative and qualitative experiments with five reader models on it show that SELD consistently outperforms strong and recent baselines across compressors, compression rates and models, identifying the evidence-removing edit in 83% of matched pairs with unchanged answers, against 70% for the strongest baseline. These results show that evidence loss in compressed contexts can be detected even when the answer is unchanged.

arXiv cs.DC13:00

Self-stabilization and Byzantine Tolerance for Maximal Independent Set

arXiv:2210.06116v4 Announce Type: replace Abstract: We analyze the impact of transient and Byzantine faults on the construction of a maximal independent set in a general network. We adapt the self-stabilizing algorithm presented by Turau for computing such a vertex set. Our algorithm is self-stabilizing and also works under the more difficult context of arbitrary Byzantine faults. Byzantine nodes can prevent nodes close to them from taking part in the independent set for an arbitrarily long time. We give boundaries to their impact by focusing on the set of all nodes excluding nodes at distance 1 or less from Byzantine nodes, and excluding some of the nodes at distance 2. As far as we know, we present the first algorithm tolerating both transient and Byzantine faults under the fair distributed daemon. We prove that this algorithm converges in $ \mathcal O(\Delta n)$ rounds w.h.p., where $n$ is the number of nodes and $\Delta$ is the maximum degree of the network. Additionally, we present a modified version of this algorithm for anonymous systems under the adversarial distributed daemon that converges in $\mathcal O(n^{2})$ moves both in expected value and with high probability.

arXiv cs.AI13:00

SetPrint: Black-Box LLM Fingerprinting via Setwise Correspondence of Response Changes

arXiv:2611.12950v1 Announce Type: new Abstract: Black-box LLM fingerprinting aims to determine whether a target model was derived from a source model when only the output responses to input queries are observable. One way to reveal this relationship is to perturb each query and observe the resulting responses of both models, yielding two corresponding sets of response variations. These sets provide two complementary sources of information: the relations among response changes within each set, and the correspondence of response changes between the two sets. Existing approaches, however, have not fully exploited both. We introduce \textbf{SetPrint}, a training-free method that jointly captures both within-set relations and between-set correspondence. SetPrint expresses each response change relative to the mean change of its own set, and then compares changes from the two models that were induced by the same perturbation. On LeaFBench, SetPrint consistently outperforms existing training-free black-box fingerprinting methods. Further analyses attribute these gains to both within-set relations and between-set correspondence.

2026-11-10

arXiv cs.CL13:00

A Model Organism for Studying Unverbalized Evaluation Awareness

arXiv:2611.09945v1 Announce Type: new Abstract: Large language models can recognize when they are being evaluated and change their behavior accordingly, undermining the reliability of safety benchmarks. Evaluation awareness (EA) is usually measured by having an LLM judge scan reasoning traces for explicit statements of recognition, but this signal is increasingly unreliable. Recent frontier-model system cards report a drop in verbalized rates between releases, which is indistinguishable from awareness that has gone unverbalized. Interpretability methods could detect such unverbalized evaluation awareness (UEA) in individual interactions, but lack instance-level ground truth. Asking the model alters the input, and dataset labels do not reveal whether it recognized a particular prompt as an evaluation. We address this gap with a model organism. Using synthetic document fine-tuning, SFT, and DPO, we train Qwen3.5-27B to emit a marker (a designated token) in its final response when it recognizes an evaluation, while suppressing the marker and verbalized awareness in its reasoning. The marker largely tracks the evaluation concept as it can be strongly modulated by evaluation framing in the system prompt. When not told whether it is being evaluated, the organism recognizes evaluations from test-question cues, mirroring what its base model often verbalizes. Using the marker as a label, we benchmark linear probes, the Jacobian Lens, and Activation Oracles, built on the base model and blind to the marker. None recovers the label reliably. These results suggest that current interpretability methods are not yet reliable UEA detectors, even when recognition relies on simple cues. Our model organism offers a testbed for improving them.