Ask or search anything...

Researchers at the University of Michigan developed a rapid numerical integrator for multi-contact systems in robotics, leveraging Event Selected Systems (ESS) theory and the Bouligand Derivative (B-Deriv) to efficiently handle simultaneous contact events without explicit root-finding. The integrator demonstrates second-order accuracy and a marginal runtime cost per contact comparable to optimized commercial simulators like MuJoCo.

An extensive international collaboration offers the first systematic review of self-evolving agents, establishing a unified theoretical framework categorized by 'what to evolve,' 'when to evolve,' and 'how to evolve'. The work consolidates diverse research, highlights key challenges, and maps applications, aiming to guide the development of AI systems capable of continuous autonomous improvement.

The Density-Calibrated Conformal Quantile Regression (CQR-d) method constructs prediction intervals that adapt to varying uncertainty across the feature space by incorporating local data density. This approach consistently maintains valid coverage while achieving substantially narrower interval widths compared to standard Conformal Quantile Regression.

Researchers from the University of Michigan and Vanderbilt University explored the effects of imposing view time limits in crowdsourced image classification tasks. The study found that a 1000ms view time limit maintained individual accuracy comparable to untimed tasks, demonstrated the robustness of consensus algorithms in preserving data quality under time pressure, and indicated worker preference for shorter limits based on psychometric scores.

Puffin, a unified multimodal model, integrates camera-centric understanding and generation by interpreting camera parameters as a "first-class modality" within a linguistic reasoning framework. The model demonstrates superior performance in both estimating camera parameters from images and generating visual content with precise viewpoint control, achieving enhanced spatial intelligence.

ALITA, developed by researchers at Princeton University and collaborating institutions, introduces a generalist AI agent that achieves leading performance on the GAIA benchmark by focusing on minimal predefinition and maximal self-evolution through dynamic Model Context Protocol (MCP) creation. The agent autonomously generates and integrates new capabilities, demonstrating that its auto-generated MCPs can also enhance smaller LLMs and other agent frameworks.

Researchers at Rutgers and the University of Michigan developed "WarAgent," an LLM-based multi-agent system to simulate historical international conflicts like WWI, WWII, and the Warring States Period. The system investigates the effectiveness of LLMs in replicating historical events, identifies crucial triggers for war, and explores the inevitability of conflicts through counterfactual analysis. GPT-4 achieved over 75% accuracy for alliances and over 90% for mobilization in anonymized WWI and WSP simulations, demonstrating that historical grievances and public morale significantly influence conflict outcomes.

THINKPRM is a generative, verbalized step-wise reward model designed to verify solutions by producing extended Chain-of-Thoughts, leveraging data-efficient fine-tuning of large reasoning models. This approach surpasses discriminative process reward models trained on significantly larger datasets, while also demonstrating robust out-of-domain generalization and improved reliability in complex reasoning tasks.

A large-scale and diverse benchmark, BIG-bench, was introduced to rigorously evaluate the capabilities and limitations of large language models across 204 tasks. The evaluation revealed that even state-of-the-art models currently achieve aggregate scores below 20 (on a 0-100 normalized scale), indicating significantly lower performance compared to human experts.

This study comprehensively characterized the cool circumgalactic medium (CGM) around galaxies at redshifts below 0.4 using data from the Dark Energy Spectroscopic Instrument (DESI) Year 1 survey. It reveals persistent correlations between cool gas absorption and galaxy properties like stellar mass and star formation rate, along with an unexpected absence of azimuthal anisotropy, indicating a possible evolution in CGM dynamics at lower redshifts.