All Articles
17909 articles total
Grounding Multi-Hop Reasoning in Structural Causal Models via Group Relative Policy Optimization
A new SCM-GRPO framework grounds multi-hop reasoning in structural causal models to reduce hallucinations in fact verification.
BioMedArena: An Open-source Toolkit for Building and Evaluating Biomedical Deep Research Agents
BioMedArena is an open-source toolkit for the standardized building and evaluation of biomedical deep research agents.
2.5-D Decomposition for LLM-Based Spatial Construction
A 2.5-D decomposition pipeline improves LLM spatial reasoning for autonomous construction by separating horizontal planning from vertical execution.
Efficient Test-time Inference for Generative Planning Models with OCL Search
The OCL search algorithm improves generative planning models by optimizing the inference process through efficient exploration control.
15 sorting algorithms in 6 minutes (2013) [video]
A classic 2013 video demonstrating 15 different sorting algorithms through visual animations.
InSight: Self-Guided Skill Acquisition via Steerable VLAs
InSight is a framework for vision-language-action (VLA) models to autonomously acquire new manipulation skills by making primitives steerable and using a VLM-guided data flywheel.
Random Rule Forest (RRF): Interpretable and Manageable Ensembles of LLM-Generated Questions for Predicting Success from Unstructured Data
Random Rule Forest (RRF) uses LLMs to generate simple YES/NO questions as weak learners for an interpretable, auditable ensemble predicting success from unstructured data.
TIP-Search: Time-Predictable Inference Scheduling for Market Prediction under Uncertain Load
TIP-Search is a time-predictable inference scheduling system for market prediction that trades accuracy for deadline satisfaction under uncertain load.
From "Aha Moments" to Controllable Thinking: Toward Meta-Cognitive Reasoning in Large Reasoning Models via Decoupled Reasoning and Control
MERA is a meta-cognitive reasoning framework that decouples reasoning from control to reduce overthinking and improve efficiency in Large Reasoning Models (LRMs).
A global log for medical AI
MedLog is a proposed universal protocol for event-level logging of medical AI interactions to enable auditing, bias detection, and performance monitoring.
Representation Interventions Enable Lifelong Knowledge Memory Control in LLMs
RILKE is a method for lifelong knowledge memory control in LLMs that uses representation-space interventions to update knowledge without full retraining.
Evolving Programmatic Skill Networks
Programmatic Skill Networks (PSN) allow agents in embodied environments to build and evolve a library of executable symbolic programs for skill reuse and adaptation.
BioPIE: A Biomedical Protocol Information Extraction Dataset for Experiment Understanding
BioPIE is a new biomedical protocol information extraction dataset providing procedure-centric knowledge graphs for better experimental understanding.
LLM-MINE: Large Language Model based Alzheimer's Disease and Related Dementias Phenotypes Mining from Clinical Notes
LLM-MINE is a framework for automatically extracting Alzheimer's disease phenotypes from unstructured clinical notes using LLMs.
Writers and Drugs
A discussion thread regarding the intersection of writers and drug use.
What I'm Finding About LLM Code Style and Token Costs
An analysis of how LLM code style and token costs vary and their implications for development.
Paying to Know: Micro-Transaction Markets for Verified Product Information in Agentic E-Commerce
Proposes a micro-transaction market for verified product information in agentic e-commerce to improve trust and competition.
Grad Detect: Gradient-Based Hallucination Detection in LLMs
Introduces Grad Detect, a gradient-based approach to detect LLM hallucinations by analyzing layer-wise gradient patterns during inference.
EG-VQA: Benchmarking Verifiable Video Question Answering with Grounded Temporal Evidence
Presents the EG-VQA benchmark for verifiable video question answering and the EG-Reasoner model for grounded reasoning.
OrbitForge: Text-to-3D Scene Generation via Reconstruction-Anchored Video Synthesis
Introduces OrbitForge, a framework that converts text-generated videos into consistent 3D Gaussian Splatting scenes.