We introduce MLE-bench, a benchmark for measuring how well AI agents perform at machine learning engineering. To this end, we curate 75 ML engineering-related competitions from Kaggle, creating a diverse set of challenging tasks that test real-world ML engineering skills such as training models, preparing datasets, and running experiments. We establish human baselines for each competition using Kaggle's publicly available leaderboards. We use open-source agent scaffolds to evaluate several frontier language models on our benchmark, finding that the best-performing setup--OpenAI's o1-preview with AIDE scaffolding--achieves at least the level of a Kaggle bronze medal in 16.9% of competitions. In addition to our main results, we investigate various forms of resource scaling for AI agents and the impact of contamination from pre-training. We open-source our benchmark code (github.com/openai/mle-bench/) to facilitate future research in understanding the ML engineering capabilities of AI agents.

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

As machine learning (ML) models find wide-spread application, there is a growing demand for interpretability: access to tools that help people see why the model made its decision. There are still many obstacles towards achieving this goal though, particularly in the context of deep learning. These obstacles stem from the scale of modern deep networks, as well as the complexity of even defining and assessing the (often context-dependent) desiderata of interpretability. Existing work on deep network interpretability has largely approached this problem from two perspectives. The first one seeks to uncover the concepts associated with specific neurons in the network, for example through visualization [Yos 15] or semantic labeling [Bau 17].

Machine learning models are often susceptible to adversarial perturbations of their inputs. Even small perturbations can cause state-of-the-art classifiers with high "standard" accuracy to produce an incorrect prediction with high confidence. To better understand this phenomenon, we study adversarially robust learning from the viewpoint of generalization. We show that already in a simple natural data model, the sample complexity of robust learning can be significantly larger than that of "standard" learning. This gap is information theoretic and holds irrespective of the training algorithm or the model family. We complement our theoretical results with experiments on popular image classification datasets and show that a similar gap exists here as well. We postulate that the difficulty of training robust classifiers stems, at least partially, from this inherently larger sample complexity.

A fundamental, and still largely unanswered, question in the context of Generative Adversarial Networks (GANs) is whether GANs are actually able to capture the key characteristics of the datasets they are trained on. The current approaches to examining this issue require significant human supervision, such as visual inspection of sampled images, and often offer only fairly limited scalability. In this paper, we propose new techniques that employ a classification-based perspective to evaluate synthetic GAN distributions and their capability to accurately reflect the essential properties of the training data. These techniques require only minimal human supervision and can easily be scaled and adapted to evaluate a variety of state-of-the-art GANs on large, popular datasets. Our analysis indicates that GANs have significant problems in reproducing the more distributional properties of the training dataset. In particular, when seen through the lens of classification, the diversity of GAN data is orders of magnitude less than that of the original data.