Large Language Models (LLMs) are increasingly asked to reason over structured data such as graphs, yet how reliably they can carry out multi-step graph algorithms in language remains unclear. Existing evaluations tend to use simple tasks on small graphs, to score code generation rather than reasoning over the graph itself, or to fix a single input format. We introduce Graph Theory Bench (GT Bench), a benchmark covering 24 classical graph problems in 44 task-structure settings, with over 100,000 examples across four representations: natural language, structured language, adjacency list, and adjacency matrix. Evaluating eight LLMs on GT Bench shows that accuracy is strongly tied to the input representation, that the best representation shifts with graph density, size, and topology as well as with the model, and that this sensitivity persists, attenuated, in the strongest reasoning models. Building on these observations, we propose the Graph Theory Agent (GTA), which pairs a preference-trained representation selector with plan-and-decompose scaffolding around a frozen executor LLM. GTA lifts Phi-4 from 53.5% to 69.1% on the benchmark's easy split and from 33.0% to 41.5% on its hard split, outperforming eight prompting and agent baselines, and transfers without retraining to GraCoRe and NLGraph. Code for benchmark generation and evaluation: https://github.com/xzx34/GTA. The project homepage is available at https://xzx34.github.io/gta/.
Zi-Xiang Xu, Yan-Bo Wang, Chenxi Wang et al.· 2 citations· ⚡1
The rapid integration of Large Vision-Language Models (VLMs) into critical infrastructure promises to revolutionize personalized healthcare and dietary management. However, in the domain of food systems, autonomous agents face a unique and persistent challenge: the"Systemic Information Asymmetry"between visual appearance and intrinsic nutritional composition. Existing benchmarks primarily focus on coarse-grained classification tasks, such as food category recognition, which fail to evaluate the intricate reasoning chain required for real-world dietary management -- specifically, the ability to traverse from identifying hidden ingredients to estimating physical mass, and finally synthesizing safety-critical medical advice. In this paper, we introduce OmniFood-Bench, a comprehensive benchmark constructed from the MM-Food-100K dataset. Unlike previous works, OmniFood-Bench evaluates VLMs across three progressive capabilities: Basic Perception (Ingredients&Cooking Methods), Quantitative Reasoning (Portion Size&Nutritional Profiling), and Safety-Critical Advisory (Disease-Specific Recommendations). We evaluate six state-of-the-art VLMs, including gpt-5.1, gemini-3-flash, and qwen3-vl-8B. Our extensive experiments reveal a startling"Semantic-Physical Gap": while models achieve near-human accuracy in naming dishes, they exhibit catastrophic failure in mass estimation and frequently hallucinate benign advice for high-risk diabetic profiles. This work establishes a rigorous standard for trustworthiness in autonomous agents deployed for public health. The code and datasets are available in: https://anonymous.4open.science/r/OmniFood-Bench-7D0B
Qian Jiang, Zhecheng Shi, Jingpu Yang et al.· 1 citation
Prior white-box studies show that large language models can retain latent traces of target knowledge after unlearning, even when the knowledge is no longer expressed in their outputs. However, existing audits remain limited to one-off diagnostics: it is unclear whether these residual signals can predict future recovery under continued training or serve as reliable optimization targets. Resolving this gap is essential to determine whether internal auditing can move beyond post-hoc evaluation toward proactive risk monitoring and safer unlearning. We propose J-Access, an inference-time audit that uses the Jacobian lens to map intermediate representations into vocabulary space and measures how often target concepts remain accessible along the model's output pathway. We hypothesize that residual accessibility reflects recovery susceptibility: knowledge that remains closer to the output pathway requires less fine-tuning to restore, leading to faster recovery. We audit 398 public unlearned models spanning eight unlearning methods. We find that: (1) most unlearned models retain access above the retain-only gold level; (2) pre-attack accessibility predicts recovery speed and extent at the model level, but cannot identify which specific facts will be recovered; and (3) directly minimizing J-Access does not promote genuine deletion. Instead, the model learns to hide knowledge from the audit, producing lower audit scores but greater post-attack recovery. These findings position J-Access as a model-level diagnostic for assessing residual susceptibility in unlearned models. We argue internal audits should serve as an independent diagnostic dimension in unlearning evaluation, and should not be converted into optimization targets without validation.
Zirui Song, Hua-Xing Liu, Xiang Wang et al.· 0 citations
Existing studies of LLM-as-judge scoring bias work predominantly at the input-output level: they perturb inputs, measure score deltas, and propose prompt-level mitigations. We argue that the same biases admit a representation-level account in the judge's hidden state, complementary to the input-output view and operationally useful in ways it does not afford. We report three findings, across seven judges, seven bias types, and nine benchmarks. Geometry: baseline judging inputs occupy a tight activation manifold while biased inputs are displaced along a low-dimensional, type-specific subspace that sharpens with depth and is recovered consistently by three families of estimators. Causal control: steering hidden states along this subspace drives scoring in both directions, forward shifts reproducing biased scoring on clean inputs and reverse shifts restoring baseline scoring on biased ones, while matched-norm random directions produce shifts an order of magnitude smaller. Operational: a simple linear projection onto the same bias-direction features anticipates judge failures on three entirely unseen benchmarks, substantially outperforming text-based alternatives. Reading bias as activation geometry, rather than as input-output noise, unifies geometric structure, causal control, and operational prediction within a single framework. The project page is available at https://xzx34.github.io/unfair-judge/
Zixiang Xu, Sixian Li, Hua-Xing Liu et al.· arXiv.org· 2 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
