We present a geometric framework for understanding Transformer-based language models, drawing an explicit analogy to General Relativity. Queries and keys induce an effective metric on representation space, and attention acts as a discrete connection that implements parallel transport of value vectors across tokens. Stacked layers provide discrete time-slices through which token representations evolve on this curved manifold, while backpropagation plays the role of a least-action principle that shapes loss-minimizing trajectories in parameter space. If this analogy is correct, token embeddings should not traverse straight paths in feature space; instead, their layer-wise steps should bend and reorient as interactions mediated by embedding space curvature. To test this prediction, we design experiments that expose both the presence and the consequences of curvature: (i) we visualize a curvature landscape for a full paragraph, revealing how local turning angles vary across tokens and layers; (ii) we show through simulations that excess counts of sharp/flat angles and longer length-to-chord ratios are not explainable by dimensionality or chance; and (iii) inspired by Einstein's eclipse experiment, we probe deflection under controlled context edits, demonstrating measurable, meaning-consistent bends in embedding trajectories that confirm attention-induced curvature.

Conversational AI is constrained in many real-world settings where only one side of a dialogue can be recorded, such as telemedicine, call centers, and smart glasses. We formalize this as the one-sided conversation problem (1SC): inferring and learning from one side of a conversation. We study two tasks: (1) reconstructing the missing speaker's turns for real-time use cases, and (2) generating summaries from one-sided transcripts. Evaluating prompting and finetuned models on MultiWOZ, DailyDialog, and Candor with both human A/B testing and LLM-as-a-judge metrics, we find that access to one future turn and information about utterance length improves reconstruction, placeholder prompting helps to mitigate hallucination, and while large models generate promising reconstructions with prompting, smaller models require finetuning. Further, high-quality summaries can be generated without reconstructing missing turns. We present 1SC as a novel challenge and report promising results that mark a step toward privacy-aware conversational AI.

Evaluating large language models (LLMs) as judges is increasingly critical for building scalable and trustworthy evaluation pipelines. We present ScalingEval, a large-scale benchmarking study that systematically compares 36 LLMs, including GPT, Gemini, Claude, and Llama, across multiple product categories using a consensus-driven evaluation protocol. Our multi-agent framework aggregates pattern audits and issue codes into ground-truth labels via scalable majority voting, enabling reproducible comparison of LLM evaluators without human annotation. Applied to large-scale complementary-item recommendation, the benchmark reports four key findings: (i) Anthropic Claude 3.5 Sonnet achieves the highest decision confidence; (ii) Gemini 1.5 Pro offers the best overall performance across categories; (iii) GPT-4o provides the most favorable latency-accuracy-cost tradeoff; and (iv) GPT-OSS 20B leads among open-source models. Category-level analysis shows strong consensus in structured domains (Electronics, Sports) but persistent disagreement in lifestyle categories (Clothing, Food). These results establish ScalingEval as a reproducible benchmark and evaluation protocol for LLMs as judges, with actionable guidance on scaling, reliability, and model family tradeoffs.

We present ROBOTO2, an open-source, web-based platform for large language model (LLM)-assisted risk of bias (ROB) assessment of clinical trials. ROBOTO2 streamlines the traditionally labor-intensive ROB v2 (ROB2) annotation process via an interactive interface that combines PDF parsing, retrieval-augmented LLM prompting, and human-in-the-loop review. Users can upload clinical trial reports, receive preliminary answers and supporting evidence for ROB2 signaling questions, and provide real-time feedback or corrections to system suggestions. ROBOTO2 is publicly available at https://roboto2.vercel.app/, with code and data released to foster reproducibility and adoption. We construct and release a dataset of 521 pediatric clinical trial reports (8954 signaling questions with 1202 evidence passages), annotated using both manually and LLM-assisted methods, serving as a benchmark and enabling future research. Using this dataset, we benchmark ROB2 performance for 4 LLMs and provide an analysis into current model capabilities and ongoing challenges in automating this critical aspect of systematic review.

Large language models (LLMs) have seen increasing popularity in enterprise applications where AI agents and humans engage in objective-driven interactions. However, these systems are difficult to evaluate: data may be complex and unlabeled; human annotation is often impractical at scale; custom metrics can monitor for specific errors, but not previously-undetected ones; and LLM judges can produce unreliable results. We introduce the first set of unsupervised metrics for objective-driven interactions, leveraging statistical properties of unlabeled interaction data and using fine-tuned LLMs to adapt to distributional shifts. We develop metrics for labeling user goals, measuring goal completion, and quantifying LLM uncertainty without grounding evaluations in human-generated ideal responses. Our approach is validated on open-domain and task-specific interaction data.

Recent work in financial machine learning has shown the virtue of complexity: the phenomenon by which deep learning methods capable of learning highly nonlinear relationships outperform simpler approaches in financial forecasting. While transformer architectures like Informer have shown promise for financial time series forecasting, the application of transformer models for options data remains largely unexplored. We conduct preliminary studies towards the development of a transformer model for options data by training the Vision Transformer (ViT) architecture, typically used in modern image recognition and classification systems, to predict the realized volatility of an asset over the next 30 days from its implied volatility surface (augmented with date information) for a single day. We show that the ViT can learn seasonal patterns and nonlinear features from the IV surface, suggesting a promising direction for model development.

Aspect-based Sentiment Analysis (ABSA) is a fine-grained opinion mining approach that identifies and classifies opinions associated with specific entities (aspects) or their categories within a sentence. Despite its rapid growth and broad potential, ABSA research and resources remain concentrated in commercial domains, leaving analytical needs unmet in high-demand yet low-resource areas such as education and healthcare. Domain adaptation challenges and most existing methods' reliance on resource-intensive in-training knowledge injection further hinder progress in these areas. Moreover, traditional evaluation methods based on exact matches are overly rigid for ABSA tasks, penalising any boundary variations which may misrepresent the performance of generative models. This work addresses these gaps through three contributions: 1) We propose a novel evaluation method, Flexible Text Similarity Matching and Optimal Bipartite Pairing (FTS-OBP), which accommodates realistic extraction boundary variations while maintaining strong correlation with traditional metrics and offering fine-grained diagnostics. 2) We present the first ABSA study of small decoder-only generative language models (SLMs; <7B parameters), examining resource lower bounds via a case study in education review ABSA. We systematically explore data-free (in-context learning and weight merging) and data-light fine-tuning methods, and propose a multitask fine-tuning strategy that significantly enhances SLM performance, enabling 1.5-3.8 B models to surpass proprietary large models and approach benchmark results with only 200-1,000 examples on a single GPU. 3) We release the first public set of education review ABSA resources to support future research in low-resource domains.

Open data repositories hold potential for evidence-based decision-making, yet are inaccessible to non-experts lacking expertise in dataset discovery, schema mapping, and statistical analysis. Large language models show promise for individual tasks, but end-to-end analytical workflows expose fundamental limitations: attention dilutes across growing contexts, specialized reasoning patterns interfere, and errors propagate undetected. We present PublicAgent, a multi-agent framework that addresses these limitations through decomposition into specialized agents for intent clarification, dataset discovery, analysis, and reporting. This architecture maintains focused attention within agent contexts and enables validation at each stage. Evaluation across five models and 50 queries derives five design principles for multi-agent LLM systems. First, specialization provides value independent of model strength--even the strongest model shows 97.5% agent win rates, with benefits orthogonal to model scale. Second, agents divide into universal (discovery, analysis) and conditional (report, intent) categories. Universal agents show consistent effectiveness (std dev 12.4%) while conditional agents vary by model (std dev 20.5%). Third, agents mitigate distinct failure modes--removing discovery or analysis causes catastrophic failures (243-280 instances), while removing report or intent causes quality degradation. Fourth, architectural benefits persist across task complexity with stable win rates (86-92% analysis, 84-94% discovery), indicating workflow management value rather than reasoning enhancement. Fifth, wide variance in agent effectiveness across models (42-96% for analysis) requires model-aware architecture design. These principles guide when and why specialization is necessary for complex analytical workflows while enabling broader access to public data through natural language interfaces.

Vision-Language Pretraining (VLP) has achieved remarkable success across various downstream tasks, but such gains are largely driven by scaling up on training data. Y et, literature methods treat image-text pairs as isolated training examples; this neglects the rich relational structure naturally present in many domains, such as e-commerce product co-purchase graphs and social recommendation networks. Inspired by neuroscientific evidence that human encodes knowledge as relationship cognitive maps, we introduce Structure-aware Language-Image Pretraining (SLIP). SLIP integrates a structural contrastive loss to align modalities while also modeling relationships between neighboring entities in a structured graph. To support this paradigm, we construct a large-scale Amazon Product Co-purchase Multi-modal Graph Dataset, enabling structured cross-modality supervision at scale. Experiment results show that SLIP consistently outperforms CLIP on cross-modal retrieval and classification tasks in both zero-shot and few-shot settings, showing the value of relational supervision for cross-modal alignment. Vision-language alignment has emerged as a key challenge in multimodal representation learning, with recent pretraining approaches achieving remarkable success by learning from web-scale data, driving progress in multimodal tasks such as image-text retrieval, visual question answering (VQA), and image captioning Gan et al. (2022). Ground-breaking work CLIP (Radford et al., 2021) has shown that a simple contrastive objective can yield state-of-the-art representations when scaled to millions of noisy image-text pairs, and such large-scale training has thus become the paradigm for vision-language foundation models. However, these web-scale corpora are notoriously noisy: captions can be generic, off-topic, or mismatched to the image.

We introduce an Analyze-Revise-Finetune (ARF) pipeline that enables smaller open-source language models (LLMs) to surpass substantially larger proprietary models in customer service summarization tasks. The pipeline first analyzes and categorizes common errors in summaries produced by a teacher model (GPT-3.5), then performs a targeted revision using a compact editor model (Llama 3.1 70B) to generate high-quality, refined training data. Fine-tuning a smaller student model (Llama 3.1 8B) on this refined data resulted in superior summarization performance compared to GPT-3.5. The ARF pipeline improves cost efficiency and data privacy while maintaining competitive accuracy, illustrating a generalizable framework for enhancing open-source LLMs across diverse downstream applications.