Science – Page 92

When Bots Behave Like Us: Modeling Social Dynamics with AI

27.01.2026 by ebaster

Through deep binding of pretrained language models with detailed backstories, coupled with temporal grounding and consistency filtering, a more robust human model emerges, capable of accurately predicting identity-driven behaviors and nuanced contextual responses.

New research shows that artificial intelligence can accurately simulate human behavior in group settings, revealing the subtle influences of identity and context on cooperation.

Collective Intelligence: Scaling Reasoning with Semantic Attention

27.01.2026 by ebaster

The Attention-MoA framework employs collaborative agents-those refining responses via semantic attention and summary agents-within each layer, while residual agents with adaptive early stopping mechanisms manage inference depth across layers, enabling dynamic control and refinement of the process.

A new framework, Attention-MoA, boosts collaborative problem-solving in large language models by enabling agents to dynamically focus on each other’s strengths.

SQL Gets a Brain Boost: Smarter Queries with Machine Learning

27.01.2026 by ebaster

The iPDB system facilitates query execution through a process designed to natively support semantic operators, enabling a more expressive and mathematically rigorous approach to data interrogation.

A new data system integrates the power of large language models directly into SQL databases to dramatically improve semantic query processing and performance.

Precision Robotics: A Single Framework for Calibrating Industrial Arms

27.01.2026 by ebaster

A new calibration method streamlines the process of achieving sub-millimeter accuracy in industrial robots by addressing multiple error sources simultaneously.

AI Writes the System: A Deep Learning Stack Built by Artificial Agents

27.01.2026 by ebaster

VibeTensor establishes a heterogeneous compute ecosystem-Python and Node.js frontends communicate with a central [latex]C++[/latex] core-where tensor operations, automatic differentiation, and CUDA runtime components are managed through shared resources and dynamically loaded extensions, anticipating future growth rather than rigid construction.

Researchers have demonstrated an AI-driven approach to creating complete deep learning systems, from user-facing Python code to optimized GPU kernels.

Smart Farming: Robots Learn to Cover More Ground with Less Energy

27.01.2026 by ebaster

An architecture combining dual convolutional layers with attention mechanisms addresses the challenge of constrained energy budgets in computation path planning, enabling efficient resource allocation during complex problem-solving.

A new reinforcement learning framework enables agricultural robots to autonomously plan energy-efficient paths for comprehensive field coverage.

Driving Intelligence: A New Benchmark for AI Agents

27.01.2026 by ebaster

The AgentDrive benchmark suite establishes a comprehensive evaluation framework-encompassing generative scenario creation ([latex]AgentDrive-Gen[/latex]), simulated outcome labeling ([latex]AgentDrive-Sim[/latex]), and rigorous reasoning assessment ([latex]AgentDrive-MCQ[/latex])-to measure the capacity of autonomous agents navigating complex driving environments.

Researchers have released a comprehensive dataset to rigorously test the reasoning and decision-making capabilities of AI systems designed for self-driving vehicles.

Remembering Users: A New Approach to Personalized Recommendations

26.01.2026 by ebaster

The system constructs a dynamic memory of user preferences and collaborative relationships by capturing behavioral history, extracting relational data to define communities, indexing these through prototype memories, and then evolving this knowledge base via simulated interactions-a process enabling the reflection and consolidation of existing information or the formation of entirely new memories.

A novel memory framework allows AI agents to better understand and predict user behavior by evolving individual experiences into collective insights.

Sharper Answers from Science: Refining AI’s Knowledge Search

26.01.2026 by ebaster

DeepEra distills complex information by initially discerning user intent, then rigorously assessing retrieved passages for mechanistic or causal relevance using large language models, ultimately delivering a concise, substantiated evidence set-a process reflecting the inevitable reduction of expansive data into manageable, meaningful form.

New research introduces a system that significantly improves the accuracy of AI-powered scientific question answering by focusing on the most relevant evidence.

AI at the Edge: Unlocking Synergistic Intelligence

26.01.2026 by ebaster

This study showcases agentic AI-RAN scenarios incorporating SC3 functionality, enabling nuanced control within low-altitude wireless networks.

A novel architecture merges sensing, communication, computing, and control at the network edge, paving the way for robust and responsive autonomous systems.