Humanity's Last Exam

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 3,000 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai.

LPLgrad: Optimizing Active Learning Through Gradient Norm Sample Selection and Auxiliary Model Training

Machine learning models are increasingly being utilized across various fields and tasks due to their outstanding performance and strong generalization capabilities. Nonetheless, their success hinges on the availability of large volumes of annotated data, the creation of which is often labor-intensive, time-consuming, and expensive. Many active learning (AL) approaches have been proposed to address these challenges, but they often fail to fully leverage the information from the core phases of AL, such as training on the labeled set and querying new unlabeled samples. To bridge this gap, we propose a novel AL approach, Loss Prediction Loss with Gradient Norm (LPLgrad), designed to quantify model uncertainty effectively and improve the accuracy of image classification tasks. LPLgrad operates in two distinct phases: (i) {\em Training Phase} aims to predict the loss for input features by jointly training a main model and an auxiliary model. Both models are trained on the labeled data to maximize the efficiency of the learning process, an aspect often overlooked in previous AL methods. This dual-model approach enhances the ability to extract complex input features and learn intrinsic patterns from the data effectively; (ii) {\em Querying Phase} that quantifies the uncertainty of the main model to guide sample selection. This is achieved by calculating the gradient norm of the entropy values for samples in the unlabeled dataset. Samples with the highest gradient norms are prioritized for labeling and subsequently added to the labeled set, improving the model's performance with minimal labeling effort. Extensive evaluations on real-world datasets demonstrate that the LPLgrad approach outperforms state-of-the-art methods by order of magnitude in terms of accuracy on a small number of labeled images, yet achieving comparable training and querying times in multiple image classification tasks.

FisherMask: Enhancing Neural Network Labeling Efficiency in Image Classification Using Fisher Information

Deep learning (DL) models are popular across various domains due to their remarkable performance and efficiency. However, their effectiveness relies heavily on large amounts of labeled data, which are often time-consuming and labor-intensive to generate manually. To overcome this challenge, it is essential to develop strategies that reduce reliance on extensive labeled data while preserving model performance. In this paper, we propose FisherMask, a Fisher information-based active learning (AL) approach that identifies key network parameters by masking them based on their Fisher information values. FisherMask enhances batch AL by using Fisher information to select the most critical parameters, allowing the identification of the most impactful samples during AL training. Moreover, Fisher information possesses favorable statistical properties, offering valuable insights into model behavior and providing a better understanding of the performance characteristics within the AL pipeline. Our extensive experiments demonstrate that FisherMask significantly outperforms state-of-the-art methods on diverse datasets, including CIFAR-10 and FashionMNIST, especially under imbalanced settings. These improvements lead to substantial gains in labeling efficiency. Hence serving as an effective tool to measure the sensitivity of model parameters to data samples. Our code is available on \url{https://github.com/sgchr273/FisherMask}.