Jeffrey Wang

Aaditya Singh, Adam Fry, Adam Perelman et al. · berkeley, mila

This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say 'think hard about this' in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. Once usage limits are reached, a mini version of each model handles remaining queries. This system card focuses primarily on gpt-5-thinking and gpt-5-main, while evaluations for other models are available in the appendix. The GPT-5 system not only outperforms previous models on benchmarks and answers questions more quickly, but -- more importantly -- is more useful for real-world queries. We've made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, and have leveled up GPT-5's performance in three of ChatGPT's most common uses: writing, coding, and health. All of the GPT-5 models additionally feature safe-completions, our latest approach to safety training to prevent disallowed content. Similarly to ChatGPT agent, we have decided to treat gpt-5-thinking as High capability in the Biological and Chemical domain under our Preparedness Framework, activating the associated safeguards. While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm -- our defined threshold for High capability -- we have chosen to take a precautionary approach.

Mehak Arora, Hassan Mortagy, Nathan Dwarshuis et al.

The objective of this work is to develop an Electronic Medical Record (EMR) data processing tool that confers clinical context to Machine Learning (ML) algorithms for error handling, bias mitigation and interpretability. We present Trust-MAPS, an algorithm that translates clinical domain knowledge into high-dimensional, mixed-integer programming models that capture physiological and biological constraints on clinical measurements. EMR data is projected onto this constrained space, effectively bringing outliers to fall within a physiologically feasible range. We then compute the distance of each data point from the constrained space modeling healthy physiology to quantify deviation from the norm. These distances, termed "trust-scores," are integrated into the feature space for downstream ML applications. We demonstrate the utility of Trust-MAPS by training a binary classifier for early sepsis prediction on data from the 2019 PhysioNet Computing in Cardiology Challenge, using the XGBoost algorithm and applying SMOTE for overcoming class-imbalance. The Trust-MAPS framework shows desirable behavior in handling potential errors and boosting predictive performance. We achieve an AUROC of 0.91 (0.89, 0.92 : 95% CI) for predicting sepsis 6 hours before onset - a marked 15% improvement over a baseline model trained without Trust-MAPS. Trust-scores emerge as clinically meaningful features that not only boost predictive performance for clinical decision support tasks, but also lend interpretability to ML models. This work is the first to translate clinical domain knowledge into mathematical constraints, model cross-vital dependencies, and identify aberrations in high-dimensional medical data. Our method allows for error handling in EMR, and confers interpretability and superior predictive power to models trained for clinical decision support.

Jeffrey Wang, Jonathan Gregory, Grigorios G. Chrysos

Modern vision backbones treat pointwise activations (e.g., ReLU, GELU) and exponential softmax as essential sources of nonlinearity, but we demonstrate they are not required within MetaFormer-style vision backbones. We design activation-free polynomial alternatives for three core primitives (MLPs, convolutions, and attention), where Hadamard products replace standard nonlinearities to yield polynomial functions of the input. These modules integrate seamlessly into existing architectures: instantiated within MetaFormer, a modular framework for vision backbones, our PolyNeXt models match or exceed activation-based counterparts across model scales on ImageNet classification, ADE20K semantic segmentation, and out-of-distribution robustness. We also substantially outperform prior polynomial networks at reduced computational cost, showing that polynomial variants of standard modules beat complex custom architectures.

Marvin Li, Jason Wang, Jeffrey Wang et al.

Recent work has shown that Large Language Models (LLMs) can unintentionally leak sensitive information present in their training data. In this paper, we present Model Perturbations (MoPe), a new method to identify with high confidence if a given text is in the training data of a pre-trained language model, given white-box access to the models parameters. MoPe adds noise to the model in parameter space and measures the drop in log-likelihood at a given point $x$, a statistic we show approximates the trace of the Hessian matrix with respect to model parameters. Across language models ranging from $70$M to $12$B parameters, we show that MoPe is more effective than existing loss-based attacks and recently proposed perturbation-based methods. We also examine the role of training point order and model size in attack success, and empirically demonstrate that MoPe accurately approximate the trace of the Hessian in practice. Our results show that the loss of a point alone is insufficient to determine extractability -- there are training points we can recover using our method that have average loss. This casts some doubt on prior works that use the loss of a point as evidence of memorization or unlearning.

Miles Wang, Tom Dupré la Tour, Olivia Watkins et al.

Understanding how language models generalize behaviors from their training to a broader deployment distribution is an important problem in AI safety. Betley et al. discovered that fine-tuning GPT-4o on intentionally insecure code causes "emergent misalignment," where models give stereotypically malicious responses to unrelated prompts. We extend this work, demonstrating emergent misalignment across diverse conditions, including reinforcement learning on reasoning models, fine-tuning on various synthetic datasets, and in models without safety training. To investigate the mechanisms behind this generalized misalignment, we apply a "model diffing" approach using sparse autoencoders to compare internal model representations before and after fine-tuning. This approach reveals several "misaligned persona" features in activation space, including a toxic persona feature which most strongly controls emergent misalignment and can be used to predict whether a model will exhibit such behavior. Additionally, we investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.

Chujie Zheng, Jeffrey Wang, Shuqian Albee Zhang et al.

We propose a novel method for evaluating the performance of a content search system that measures the semantic match between a query and the results returned by the search system. We introduce a metric called "on-topic rate" to measure the percentage of results that are relevant to the query. To achieve this, we design a pipeline that defines a golden query set, retrieves the top K results for each query, and sends calls to GPT 3.5 with formulated prompts. Our semantic evaluation pipeline helps identify common failure patterns and goals against the metric for relevance improvements.