ToolAlignBench: Investigating Alignment Conflicts in Tool-Calling Enabled LLMs

Safety alignment in LLMs aims to align models with human values, but which values take precedence when they conflict? We investigate this question in the context of tool-calling LLM agents deployed in regulated industries, where agents processing confidential documents may encounter content that triggers safety-trained values (e.g., public welfare) that conflict with deployment-context instructions (e.g., internal logging). To empirically verify this phenomenon, we build a benchmark of 128 scenarios across 16 domains. We find that safety-aligned open-source models override their deployment instructions up to 43.4% of the time, engaging in whistleblowing, data exfiltration, and evidence tampering when processing documents that suggest organizational wrongdoing. We also find that abliteration reduces rates of external whistleblowing. These results reveal a fundamental tension in pluralistic alignment, where the same safety training that protects users can cause agents to act against deployment instructions in ways that create unpredictable liability risks. We release our benchmark as a framework to support evaluation of agent behavior under competing legitimate interests.

When Does Personality Composition Matter for Multi-Agent LLM Teams?

Personality prompting shapes how large language models communicate, yet whether these behavioral shifts affect objective task outcomes remains under-explored. Prior work shows that agents prompted with low agreeableness produce adversarial language, while those prompted with high agreeableness become cooperative, but the relationship between communication style and task performance has not been systematically examined across multiple domains. In this work, we investigate whether personality composition matters for multi-agent team performance by manipulating personality traits across frontier LLMs on three task domains: structured coding, open-ended research collaboration, and competitive bargaining. We find that personality effects depend critically on task structure. In coding tasks, low agreeableness leads to large communication shifts that have little effect on milestone completion. In open-ended collaboration and bargaining, the same manipulation substantially degrades performance. We discuss implications for multi-agent system design and the limits of personality manipulation.

Tri-Accel: Curvature-Aware Precision-Adaptive and Memory-Elastic Optimization for Efficient GPU Usage

Deep neural networks are increasingly bottlenecked by the cost of optimization, both in terms of GPU memory and compute time. Existing acceleration techniques, such as mixed precision, second-order methods, and batch size scaling, are typically used in isolation. We present Tri-Accel, a unified optimization framework that co-adapts three acceleration strategies along with adaptive parameters during training: (1) Precision-Adaptive Updates that dynamically assign mixed-precision levels to layers based on curvature and gradient variance; (2) Sparse Second-Order Signals that exploit Hessian/Fisher sparsity patterns to guide precision and step size decisions; and (3) Memory-Elastic Batch Scaling that adjusts batch size in real time according to VRAM availability. On CIFAR-10 with ResNet-18 and EfficientNet-B0, Tri-Accel achieves up to 9.9% reduction in training time and 13.3% lower memory usage, while improving accuracy by +1.1 percentage points over FP32 baselines. Tested on CIFAR-10/100, our approach demonstrates adaptive learning behavior, with efficiency gradually improving over the course of training as the system learns to allocate resources more effectively. Compared to static mixed-precision training, Tri-Accel maintains 78.1% accuracy while reducing memory footprint from 0.35GB to 0.31GB on standard hardware. The framework is implemented with custom Triton kernels, whose hardware-aware adaptation enables automatic optimization without manual hyperparameter tuning, making it practical for deployment across diverse computational environments. This work demonstrates how algorithmic adaptivity and hardware awareness can be combined to improve scalability in resource-constrained settings, paving the way for more efficient neural network training on edge devices and cost-sensitive cloud deployments.

Do LLMs Understand Ambiguity in Text? A Case Study in Open-world Question Answering

Ambiguity in natural language poses significant challenges to Large Language Models (LLMs) used for open-domain question answering. LLMs often struggle with the inherent uncertainties of human communication, leading to misinterpretations, miscommunications, hallucinations, and biased responses. This significantly weakens their ability to be used for tasks like fact-checking, question answering, feature extraction, and sentiment analysis. Using open-domain question answering as a test case, we compare off-the-shelf and few-shot LLM performance, focusing on measuring the impact of explicit disambiguation strategies. We demonstrate how simple, training-free, token-level disambiguation methods may be effectively used to improve LLM performance for ambiguous question answering tasks. We empirically show our findings and discuss best practices and broader impacts regarding ambiguity in LLMs.