The Structural Safety Generalization Problem

LLM jailbreaks are a widespread safety challenge. Given this problem has not yet been tractable, we suggest targeting a key failure mechanism: the failure of safety to generalize across semantically equivalent inputs. We further focus the target by requiring desirable tractability properties of attacks to study: explainability, transferability between models, and transferability between goals. We perform red-teaming within this framework by uncovering new vulnerabilities to multi-turn, multi-image, and translation-based attacks. These attacks are semantically equivalent by our design to their single-turn, single-image, or untranslated counterparts, enabling systematic comparisons; we show that the different structures yield different safety outcomes. We then demonstrate the potential for this framework to enable new defenses by proposing a Structure Rewriting Guardrail, which converts an input to a structure more conducive to safety assessment. This guardrail significantly improves refusal of harmful inputs, without over-refusing benign ones. Thus, by framing this intermediate challenge - more tractable than universal defenses but essential for long-term safety - we highlight a critical milestone for AI safety research.

From Intuition to Understanding: Using AI Peers to Overcome Physics Misconceptions

Generative AI has the potential to transform personalization and accessibility of education. However, it raises serious concerns about accuracy and helping students become independent critical thinkers. In this study, we designed a helpful AI "Peer" to help students correct fundamental physics misconceptions related to Newtonian mechanic concepts. In contrast to approaches that seek near-perfect accuracy to create an authoritative AI tutor or teacher, we directly inform students that this AI can answer up to 40% of questions incorrectly. In a randomized controlled trial with 165 students, those who engaged in targeted dialogue with the AI Peer achieved post-test scores that were, on average, 10.5 percentage points higher - with over 20 percentage points higher normalized gain - than a control group that discussed physics history. Qualitative feedback indicated that 91% of the treatment group's AI interactions were rated as helpful. Furthermore, by comparing student performance on pre- and post-test questions about the same concept, along with experts' annotations of the AI interactions, we find initial evidence suggesting the improvement in performance does not depend on the correctness of the AI. With further research, the AI Peer paradigm described here could open new possibilities for how we learn, adapt to, and grow with AI.

PairBench: A Systematic Framework for Selecting Reliable Judge VLMs

As large vision language models (VLMs) are increasingly used as automated evaluators, understanding their ability to effectively compare data pairs as instructed in the prompt becomes essential. To address this, we present PairBench, a low-cost framework that systematically evaluates VLMs as customizable similarity tools across various modalities and scenarios. Through PairBench, we introduce four metrics that represent key desiderata of similarity scores: alignment with human annotations, consistency for data pairs irrespective of their order, smoothness of similarity distributions, and controllability through prompting. Our analysis demonstrates that no model, whether closed- or open-source, is superior on all metrics; the optimal choice depends on an auto evaluator's desired behavior (e.g., a smooth vs. a sharp judge), highlighting risks of widespread adoption of VLMs as evaluators without thorough assessment. For instance, the majority of VLMs struggle with maintaining symmetric similarity scores regardless of order. Additionally, our results show that the performance of VLMs on the metrics in PairBench closely correlates with popular benchmarks, showcasing its predictive power in ranking models.

Higher Order Transformers: Enhancing Stock Movement Prediction On Multimodal Time-Series Data

In this paper, we tackle the challenge of predicting stock movements in financial markets by introducing Higher Order Transformers, a novel architecture designed for processing multivariate time-series data. We extend the self-attention mechanism and the transformer architecture to a higher order, effectively capturing complex market dynamics across time and variables. To manage computational complexity, we propose a low-rank approximation of the potentially large attention tensor using tensor decomposition and employ kernel attention, reducing complexity to linear with respect to the data size. Additionally, we present an encoder-decoder model that integrates technical and fundamental analysis, utilizing multimodal signals from historical prices and related tweets. Our experiments on the Stocknet dataset demonstrate the effectiveness of our method, highlighting its potential for enhancing stock movement prediction in financial markets.

Higher Order Transformers: Efficient Attention Mechanism for Tensor Structured Data

Transformers are now ubiquitous for sequence modeling tasks, but their extension to multi-dimensional data remains a challenge due to the quadratic cost of the attention mechanism. In this paper, we propose Higher-Order Transformers (HOT), a novel architecture designed to efficiently process data with more than two axes, i.e. higher-order tensors. To address the computational challenges associated with high-order tensor attention, we introduce a novel Kronecker factorized attention mechanism that reduces the attention cost to quadratic in each axis' dimension, rather than quadratic in the total size of the input tensor. To further enhance efficiency, HOT leverages kernelized attention, reducing the complexity to linear. This strategy maintains the model's expressiveness while enabling scalable attention computation. We validate the effectiveness of HOT on two high-dimensional tasks, including multivariate time series forecasting, and 3D medical image classification. Experimental results demonstrate that HOT achieves competitive performance while significantly improving computational efficiency, showcasing its potential for tackling a wide range of complex, multi-dimensional data.

Epistemic Integrity in Large Language Models

Large language models are increasingly relied upon as sources of information, but their propensity for generating false or misleading statements with high confidence poses risks for users and society. In this paper, we confront the critical problem of epistemic miscalibration $\unicode{x2013}$ where a model's linguistic assertiveness fails to reflect its true internal certainty. We introduce a new human-labeled dataset and a novel method for measuring the linguistic assertiveness of Large Language Models (LLMs) which cuts error rates by over 50% relative to previous benchmarks. Validated across multiple datasets, our method reveals a stark misalignment between how confidently models linguistically present information and their actual accuracy. Further human evaluations confirm the severity of this miscalibration. This evidence underscores the urgent risk of the overstated certainty LLMs hold which may mislead users on a massive scale. Our framework provides a crucial step forward in diagnosing this miscalibration, offering a path towards correcting it and more trustworthy AI across domains.

A Guide to Misinformation Detection Datasets

Misinformation is a complex societal issue, and mitigating solutions are difficult to create due to data deficiencies. To address this problem, we have curated the largest collection of (mis)information datasets in the literature, totaling 75. From these, we evaluated the quality of all of the 36 datasets that consist of statements or claims. We assess these datasets to identify those with solid foundations for empirical work and those with flaws that could result in misleading and non-generalizable results, such as insufficient label quality, spurious correlations, or political bias. We further provide state-of-the-art baselines on all these datasets, but show that regardless of label quality, categorical labels may no longer give an accurate evaluation of detection model performance. We discuss alternatives to mitigate this problem. Overall, this guide aims to provide a roadmap for obtaining higher quality data and conducting more effective evaluations, ultimately improving research in misinformation detection. All datasets and other artifacts are available at https://misinfo-datasets.complexdatalab.com/.

Hallucination Detox: Sensitive Neuron Dropout (SeND) for Large Language Model Training

As large language models (LLMs) become increasingly deployed across various industries, concerns regarding their reliability, particularly due to hallucinations-outputs that are factually inaccurate or irrelevant to user input-have grown. Our research investigates the relationship between the training process and the emergence of hallucinations to address a key gap in existing research that focuses primarily on post hoc detection and mitigation strategies. Using models from the Pythia suite (70M-12B parameters) and several hallucination detection metrics, we analyze hallucination trends throughout training and explore LLM internal dynamics. We introduce SEnsitive Neuron Dropout (SeND), a novel training protocol designed to mitigate hallucinations by reducing variance during training. SeND achieves this by deterministically dropping neurons with significant variability on a dataset, referred to as Sensitive Neurons. In addition, we develop an unsupervised hallucination detection metric, Efficient EigenScore (EES), which approximates the traditional EigenScore in 2x speed. This efficient metric is integrated into our protocol, allowing SeND to be both computationally scalable and effective at reducing hallucinations. Our empirical evaluation demonstrates that our approach improves LLM reliability at test time by up to 40% compared to normal training while also providing an efficient method to improve factual accuracy when adapting LLMs to domains such as Wikipedia and Medical datasets.

A Simulation System Towards Solving Societal-Scale Manipulation

The rise of AI-driven manipulation poses significant risks to societal trust and democratic processes. Yet, studying these effects in real-world settings at scale is ethically and logistically impractical, highlighting a need for simulation tools that can model these dynamics in controlled settings to enable experimentation with possible defenses. We present a simulation environment designed to address this. We elaborate upon the Concordia framework that simulates offline, `real life' activity by adding online interactions to the simulation through social media with the integration of a Mastodon server. We improve simulation efficiency and information flow, and add a set of measurement tools, particularly longitudinal surveys. We demonstrate the simulator with a tailored example in which we track agents' political positions and show how partisan manipulation of agents can affect election results.

Emerging Vulnerabilities in Frontier Models: Multi-Turn Jailbreak Attacks

Large language models (LLMs) are improving at an exceptional rate. However, these models are still susceptible to jailbreak attacks, which are becoming increasingly dangerous as models become increasingly powerful. In this work, we introduce a dataset of jailbreaks where each example can be input in both a single or a multi-turn format. We show that while equivalent in content, they are not equivalent in jailbreak success: defending against one structure does not guarantee defense against the other. Similarly, LLM-based filter guardrails also perform differently depending on not just the input content but the input structure. Thus, vulnerabilities of frontier models should be studied in both single and multi-turn settings; this dataset provides a tool to do so.