All Articles
17646 articles total
Text-Driven 3D Indoor Scene Synthesis in Non-Manhattan Environments
Introduces SPG-Layout, a framework for generating physically plausible 3D indoor scenes in complex, non-Manhattan (non-orthogonal) environments.
Automated grading of Linux/bash examinations using large language models: a four-level cognitive taxonomy approach
Explores the use of LLMs to automate the grading of Linux/bash examinations using a four-level cognitive taxonomy.
EvoPolicyGym: Evaluating Autonomous Policy Evolution in Interactive Environments
Introduces EvoPolicyGym, a benchmark to evaluate how autonomous agents iteratively improve executable policies through interaction feedback.
G-RRM: Guiding Symbolic Solvers with Recurrent Reasoning Models
Proposes G-RRM, a neuro-symbolic approach that uses Recurrent Reasoning Models to guide symbolic solvers like Glucose and CaDiCaL for faster constraint satisfaction.
What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates
Studies 'public vs. private' divergence in multi-agent LLM debates, finding that agents often express different views in off-the-record channels.
Alibaba to ban Claude Code in workplace over alleged backdoor risks, source says
Alibaba is reportedly planning to ban the use of Claude Code in its workplace due to concerns over potential backdoor risks.
Quake in 13 Kilobytes (2021)
A technical demonstration of Quake implemented in just 13 Kilobytes, showcasing extreme code optimization.
A rubric-based controlled comparison of frontier language models on expert-authored clinical reasoning tasks
A study evaluating frontier LLMs on complex clinical reasoning tasks using a rubric-based approach, finding that models struggle with high-criticality criteria.
UA-ChatDev: Uncertainty-Aware Multi-Agent Collaboration for Reliable Software Development
Introduction of UA-ChatDev, a multi-agent software development framework that uses uncertainty quantification to reduce hallucination propagation.
Criticality-Based Guard Rail Validation for AI Agent Decisions in Autonomous Telecom Networks
Proposed Guard Rail Validation (GRV) framework to intercept and validate AI agent decisions in autonomous telecom networks for safety and compliance.
Purified OPSD: On-Policy Self-Distillation Without Losing How to Think
Purified OPSD proposes a two-step solution to prevent rote memorization in long-CoT reasoning models during on-policy self-distillation.
Copewell: A Multi-Agent Swarm Architecture for Equitable Mental Wellness Support
Copewell is a multi-agent swarm architecture designed to provide equitable mental wellness support through human-centered AI.
AgenticSTS: A Bounded-Memory Testbed for Long-Horizon LLM Agents
AgenticSTS introduces a bounded-memory testbed using Slay the Spire 2 to study how explicit memory layers affect long-horizon LLM agent decisions.
A Hippocampus for Linear Attention: An Exact Memory for What the Recurrent State Forgets
HOLA (Hippocampal Linear Attention) adds a bounded exact KV cache to linear attention models to improve memory and retrieval performance.
Grounded autonomous research: a fault-tolerant LLM pipeline from corpus to manuscript in frontier computational physics
A fault-tolerant LLM pipeline for autonomous research in computational physics that grounds findings in external literature to prevent hallucinations.
How working with a blind client revealed invisible accessibility gaps
A personal account of how working with a blind client highlighted accessibility gaps in software and the importance of inclusive design.
AI is 'not smart' so what's next in artificial intelligence?
A discussion on the perceived limitations of current AI and speculation on the next evolutionary step in artificial intelligence.
PACE: A Proxy for Agentic Capability Evaluation
Introduces PACE, a framework that uses a small subset of non-agentic benchmarks to accurately and cheaply predict the performance of LLM agents on expensive benchmarks.
Algebraic Model Counting for Global Analysis of Optimal Decision Trees
Proposes Algebraic Decision Tree Counting (ADTC), a formal framework for exhaustive analysis of optimal decision trees to improve explainability in AI.
Evidence-State Rewards for Long-Context Reasoning
Presents Maven, an RL framework featuring an editable evidence memory to improve long-context reasoning by rewarding stateful evidence navigation.