Do Not Trust the Benchmark: Limitations of General LLM Rankings and a Case for Task-Specific Evaluation
1mês
- Publicado
- Coletado
Autores
Danial Amin
Categorias
Benchmark scores increasingly influence the development, marketing, and selection of large language models (LLMs). Yet an overall score is interpretable only in relation to the system tested, the questions included, and the conditions of evaluation. This perspective examines five connected limitations of general LLM rankings: differences between evaluated and publicly available systems; commercial incentives and dependencies in external evaluation; benchmark saturation, defective tests, and data contamination; models exploiting scoring procedures; and the limited relevance of general scores to users' tasks. Documented cases illustrate why these problems require different responses. I argue for evaluation procedures that disclose the tested configuration, validate questions and successful task completion, report performance alongside cost and execution time, and make the scope of generalization explicit. I then discuss Isotanta, a crowdsourced benchmarking platform, as a practical example of contributed questions and repeated evaluation. A larger question pool may improve task coverage, while repeated sampling can improve the stability of estimates on that pool; neither guarantees validity or personalization. The paper distinguishes the platform's current shared ranking from proposed task-specific and user-provided evaluations. Its central argument is that model selection requires evidence about performance on the intended work, not simply a high position on a general leaderboard.
O overfeed.news indexa e aponta. Publicamos um trecho curto — o artigo completo fica em arXiv AI Papers.
Mais de arXiv AI Papers
Entre para seguir esta fonte
Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success
Robotic foundation models are typically evaluated through costly closed-loop rollouts, which motivates cheap offline proxies for model capability. The Platonic Representation Hypothesis (PRH) suggests that as representations improve, different modalities such as vision and language converge toward a shared statistical model of reality. We ask whether motion trajectories form another such modality: as embodied models become more capable, should their sensory representations converge toward the structure of physical motion? To examine this, we propose Sensorimotor Alignment (S^2), a training-free offline metric that measures how well a model's sensory representation aligns with the geometry of physical motion. We find that S^2 correlates positively with downstream robot task success across embodied model families, making it a low-cost offline proxy for prioritizing models before closed-loop evaluation. S^2 can also identify promising vision-language model (VLM) backbones even before costly fine-tuning into vision-language-action models (VLAs). Beyond this correlation, S^2 offers representation-level explanations for phenomena observed in VLAs and world action models (WAMs). With S^2, we extend PRH to embodied models and make sensorimotor alignment empirically measurable, offering insight into why an embodied model succeeds beyond its closed-loop success rate.
Direct Pass@k Optimization for Thought Pruning: Balancing Quality and Diversity in Test-Time Reasoning
Test-time scaling has emerged as a key inference strategy for reasoning models, where a model generates a large number of candidate solutions in the hope of finding at least one correct solution. Running all candidate solutions to completion can be computationally demanding, so recent work prunes the candidate pool after generating only a short prefix, committing further computation only to the surviving candidates. Existing pruning strategies divide broadly into two categories: Quality-based methods assign a score to each partial solution and discard those predicted to ultimately lead to an incorrect final solution, while diversity-based methods select partial solutions that follow different reasoning trajectories and are therefore likely to cover different final solutions. The tension between these two objectives is an instance of exploration versus exploitation, yet little to no prior work on pruning exposes this trade-off directly, let alone optimizes it. In this paper, we introduce a pruning procedure that addresses this trade-off directly by optimizing pass@k over the selected solution pool. In particular, it chooses the pool to maximize the probability that at least one of its partial solutions, when continued to completion, yields a correct answer. Because this objective simultaneously rewards high-quality partial solutions and penalizes redundancy among them, quality and diversity are naturally balanced by the criterion itself rather than through separate ad-hoc design choices. On eight open-weight reasoning models, our method outperforms existing quality-based and diversity-based pruning strategies. The gains are particularly pronounced when the pruning procedure keeps a moderate number of partial solutions for continued generation, highlighting the value of directly balancing quality and diversity during pruning.
Phase Transition from Random Features to Fully Trainable Networks
Constructing accurate and computationally tractable approximation spaces is a fundamental task in scientific computing. Partially trainable neural networks bridge fixed random features and fully adaptive representations. This paper quantifies a phase transition pattern in approximation bounds and observes analogous behavior in training errors: a rapid but smooth initial improvement followed by diminishing returns, with a sharper transition under greater initialization mismatch. Under polynomial spectral decay with exponent α>1 and leverage compatibility, we establish a squared L_2 approximation rate {O}((κ_P(m/δ)/m)^α) for two-layer networks of width m and trainable proportion P, where κ_P=(P/κ_1^*+(1-P)/κ_0^*)^{-1}. The bound holds uniformly over the unit ball of a reproducing kernel Hilbert space (RKHS) with probability at least 1-δ. The constants κ_0^* and κ_1^* quantify the fixed and trainable groups' sampling mismatches relative to the target RKHS. For spherical biased-ReLU features in dimension d, we verify these assumptions with α=(d+2)/(d-1) and κ_0^*=4{2}τ(d+1)/d for initialization scale τ1, and κ_1^*=1 under ridge leverage sampling. For d=5 and τ=10, a trainable proportion of approximately 4.1\% captures 90\% of the reduction in the approximation bound from P=0 to P=1. A structured three-layer example illustrates how a few trainable first-layer neurons exploit fixed downstream blocks through nonlinear composition. Numerical experiments across diverse settings, including networks with up to 15 layers, exhibit similar phase transitions in training errors.
Reliable AI-Assisted Decisions via Selective Auditing
We consider online monitoring of an AI system that produces a sequence of decisions. At each round, the system outputs a proposed decision together with a confidence score estimating the trustworthiness of the decision. The decision and confidence score need not be produced by the same model: for example, one agent may propose a decision, while a separate verifier agent evaluates it and outputs a correctness or safety score. Based on this information, we must decide whether to accept the AI decision automatically or consult an expert, such as a human auditor, who reviews the proposal and either accepts or corrects it. We formulate this task as an online constrained optimization problem in which the goal is to minimize the frequency of expert auditing while guaranteeing that the fraction of AI errors that go unaudited stays below a user-defined threshold. We derive a family of algorithms with provable error-control guarantees with no assumptions, including a method that is optimally efficient: it asymptotically attains the best achievable auditing rate for both adversarial and i.i.d. data streams. Experiments on screening harmful requests to large language models, qualitative text annotation, and data labeling show that our methods achieve valid error control in practice while substantially reducing the need for human auditing compared to baselines.