Victoria R. Li * Harvard University Johnathan L. Sun * Harvard University Martin Wattenberg Harvard University, Google ResearchFigure 1: We test two prominent vision-language models, GPT 4.1 and Claude 3.5, on three settings motivated by common data analysis tasks. In each setting, we ask models to describe datasets under five conditions, providing: (1) numerical data alone, (2,3) numerical data with incorrect or blank visuals as baselines, (4) numerical data with correct visuals, and (5) correct visuals alone. After 12,000 trials, we find consistent performance gains when models are provided with accurate visualizations. A BSTRACT Charts and graphs help people analyze data, but can they also be useful to AI systems? To investigate this question, we perform a series of experiments with two commercial vision-language models: GPT 4.1 and Claude 3.5. Across three representative analysis tasks, the two systems describe synthetic datasets more precisely and accurately when raw data is accompanied by a scatterplot, especially as datasets grow in complexity. Comparison with two baselines-- providing a blank chart and a chart with mismatched data--shows that the improved performance is due to the content of the charts.

State-of-the-art (SOTA) image and text generation models are multimodal models that have many similarities to large language models (LLMs). Despite achieving strong performances, leading foundational multimodal model architectures frequently lag behind the architectural sophistication of contemporary LLMs. We propose GRR-CoCa, an improved SOTA Contrastive Captioner (CoCa) model that incorporates Gaussian error gated linear units, root mean squared normalization, and rotary positional embedding into the textual decoders and the vision transformer (ViT) encoder. Each architectural modification has been shown to improve model performance in LLMs, but has yet to be adopted in CoCa. We benchmarked GRR-CoCa against Baseline CoCa, a model with the same modified textual decoders but with CoCa's original ViT encoder. We used standard pretraining and fine-tuning workflows to benchmark the models on contrastive and generative tasks. Our GRR-CoCa significantly outperformed Baseline CoCa on the pretraining dataset and three diverse fine-tuning datasets. Pretraining improvements were 27.25% in contrastive loss, 3.71% in perplexity, and 7.15% in CoCa loss. The average fine-tuning improvements were 13.66% in contrastive loss, 5.18% in perplexity, and 5.55% in CoCa loss. We show that GRR-CoCa's modified architecture improves performance and generalization across vision-language domains.

Identifying the factors driving data shifts in tabular datasets is a significant challenge for analysis and decision support systems, especially those focusing on healthcare. Privacy rules restrict data access, and noise from complex processes hinders analysis. To address this challenge, we propose SIFOTL (Statistically-Informed Fidelity-Optimization Method for Tabular Learning) that (i) extracts privacy-compliant data summary statistics, (ii) employs twin XGBoost models to disentangle intervention signals from noise with assistance from LLMs, and (iii) merges XGBoost outputs via a Pareto-weighted decision tree to identify interpretable segments responsible for the shift. Unlike existing analyses which may ignore noise or require full data access for LLM-based analysis, SIFOTL addresses both challenges using only privacy-safe summary statistics. Demonstrating its real-world efficacy, for a MEPS panel dataset mimicking a new Medicare drug subsidy, SIFOTL achieves an F1 score of 0.85, substantially outperforming BigQuery Contribution Analysis (F1=0.46) and statistical tests (F1=0.20) in identifying the segment receiving the subsidy. Furthermore, across 18 diverse EHR datasets generated based on Synthea ABM, SIFOTL sustains F1 scores of 0.86-0.96 without noise and >= 0.75 even with injected observational noise, whereas baseline average F1 scores range from 0.19-0.67 under the same tests. SIFOTL, therefore, provides an interpretable, privacy-conscious workflow that is empirically robust to observational noise.

Do larger and more capable language models learn to update their "beliefs" about propositions more consistently with Bayes' theorem when presented with evidence in-context? To test this, we formulate a Bayesian Coherence Coefficient (BCC) metric and generate a dataset with which to measure the BCC. We measure BCC for multiple pre-trained-only language models across five model families, comparing against the number of model parameters, the amount of training data, and model scores on common benchmarks. Our results provide evidence for our hypothesis that larger and more capable pre-trained language models assign credences that are more coherent with Bayes' theorem. These results have important implications for our understanding and governance of LLMs.

Large language models (LLMs) present intriguing opportunities to enhance user interaction with traditional algorithms and tools in real-world applications. An advanced planning system (APS) is a sophisticated software that leverages optimization to help operations planners create, interpret, and modify an operational plan. While highly beneficial, many customers are priced out of using an APS due to the ongoing costs of consultants responsible for customization and maintenance. To address the need for a more accessible APS expressed by supply chain planners, we present SmartAPS, a conversational system built on a tool-augmented LLM. Our system provides operations planners with an intuitive natural language chat interface, allowing them to query information, perform counterfactual reasoning, receive recommendations, and execute scenario analysis to better manage their operation. A short video demonstrating the system has been released: https://youtu.be/KtIrJjlDbyw

Text-to-image (T2I) models have become prevalent across numerous applications, making their robust evaluation against adversarial attacks a critical priority. Continuous access to new and challenging adversarial prompts across diverse domains is essential for stress-testing these models for resilience against novel attacks from multiple vectors. Current techniques for generating such prompts are either entirely authored by humans or synthetically generated. On the one hand, datasets of human-crafted adversarial prompts are often too small in size and imbalanced in their cultural and contextual representation. On the other hand, datasets of synthetically-generated prompts achieve scale, but typically lack the realistic nuances and creative adversarial strategies found in human-crafted prompts. To combine the strengths of both human and machine approaches, we propose Seed2Harvest, a hybrid red-teaming method for guided expansion of culturally diverse, human-crafted adversarial prompt seeds. The resulting prompts preserve the characteristics and attack patterns of human prompts while maintaining comparable average attack success rates (0.31 NudeNet, 0.36 SD NSFW, 0.12 Q16). Our expanded dataset achieves substantially higher diversity with 535 unique geographic locations and a Shannon entropy of 7.48, compared to 58 locations and 5.28 entropy in the original dataset. Our work demonstrates the importance of human-machine collaboration in leveraging human creativity and machine computational capacity to achieve comprehensive, scalable red-teaming for continuous T2I model safety evaluation.

Application systems using natural language interfaces to databases (NLIDBs) have democratized data analysis. This positive development has also brought forth an urgent challenge to help users who might use these systems without a background in statistical analysis to formulate bias-free analytical questions. Although significant research has focused on text-to-SQL generation accuracy, addressing cognitive biases in analytical questions remains underexplored. We present VeriMinder, https://veriminder.ai, an interactive system for detecting and mitigating such analytical vulnerabilities. Our approach introduces three key innovations: (1) a contextual semantic mapping framework for biases relevant to specific analysis contexts (2) an analytical framework that operationalizes the Hard-to-Vary principle and guides users in systematic data analysis (3) an optimized LLM-powered system that generates high-quality, task-specific prompts using a structured process involving multiple candidates, critic feedback, and self-reflection. User testing confirms the merits of our approach. In direct user experience evaluation, 82.5% participants reported positively impacting the quality of the analysis. In comparative evaluation, VeriMinder scored significantly higher than alternative approaches, at least 20% better when considered for metrics of the analysis's concreteness, comprehensiveness, and accuracy. Our system, implemented as a web application, is set to help users avoid "wrong question" vulnerability during data analysis. VeriMinder code base with prompts, https://reproducibility.link/veriminder, is available as an MIT-licensed open-source software to facilitate further research and adoption within the community.

Recent advances in agentic systems for data analysis have emphasized automation of insight generation through multi-agent frameworks, and orchestration layers. While these systems effectively manage tasks like query translation, data transformation, and visualization, they often overlook the structured reasoning process underlying analytical thinking. Reasoning large language models (LLMs) used for multi-step problem solving are trained as general-purpose problem solvers. As a result, their reasoning or thinking steps do not adhere to fixed processes for specific tasks. Real-world data analysis requires a consistent cognitive workflow: interpreting vague goals, grounding them in contextual knowledge, constructing abstract plans, and adapting execution based on intermediate outcomes. We introduce I2I-STRADA (Information-to-Insight via Structured Reasoning Agent for Data Analysis), an agentic architecture designed to formalize this reasoning process. I2I-STRADA focuses on modeling how analysis unfolds via modular sub-tasks that reflect the cognitive steps of analytical reasoning. Evaluations on the DABstep and DABench benchmarks show that I2I-STRADA outperforms prior systems in planning coherence and insight alignment, highlighting the importance of structured cognitive workflows in agent design for data analysis.

Recent advances in text-to-image (T2I) generation have led to impressive visual results. However, these models still face significant challenges when handling complex prompt, particularly those involving multiple subjects with distinct attributes. Inspired by the human drawing process, which first outlines the composition and then incrementally adds details, we propose Detail++, a training-free framework that introduces a novel Progressive Detail Injection (PDI) strategy to address this limitation. Specifically, we decompose a complex prompt into a sequence of simplified sub-prompts, guiding the generation process in stages. This staged generation leverages the inherent layout-controlling capacity of self-attention to first ensure global composition, followed by precise refinement. To achieve accurate binding between attributes and corresponding subjects, we exploit cross-attention mechanisms and further introduce a Centroid Alignment Loss at test time to reduce binding noise and enhance attribute consistency. Extensive experiments on T2I-CompBench and a newly constructed style composition benchmark demonstrate that Detail++ significantly outperforms existing methods, particularly in scenarios involving multiple objects and complex stylistic conditions.

Process Reward Models (PRMs) are crucial for guiding Large Language Models (LLMs) in complex scenarios by providing dense reward signals. However, existing PRMs primarily rely on heuristic approaches, which struggle with cross-domain generalization. While LLM-as-judge has been proposed to provide generalized rewards, current research has focused mainly on feedback results, overlooking the meaningful guidance embedded within the text. Additionally, static and coarse-grained evaluation criteria struggle to adapt to complex process supervision. To tackle these challenges, we propose Dynamic and Generalizable Process Reward Modeling (DG-PRM), which features a reward tree to capture and store fine-grained, multi-dimensional reward criteria. DG-PRM dynamically selects reward signals for step-wise reward scoring. To handle multifaceted reward signals, we pioneeringly adopt Pareto dominance estimation to identify discriminative positive and negative pairs. Experimental results show that DG-PRM achieves stunning performance on prevailing benchmarks, significantly boosting model performance across tasks with dense rewards. Further analysis reveals that DG-PRM adapts well to out-of-distribution scenarios, demonstrating exceptional generalizability.