All notes, newest first
2,343 notes. Notes 201 to 250, page 5 of 47.
MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use
Researchers have introduced a new benchmark called MemTrapBench to test the performance of large language models (LLMs) in terms of cognitive traps caused by their memory components. The study identifies two types of…
MidTool: Mid-training Data Synthesis for Agentic Tool Use
Researchers have developed MidTool, a system for teaching large language models how to use tools effectively during mid-training. This is done by combining large-scale web and PDF data with synthetic supervision from…
MidTool: Mid-training Data Synthesis for Agentic Tool Use
In this research, the authors introduce MidTool, a system used for training large language models during the mid-training stage. MidTool focuses on improving the model's abilities in general tool use, which includes…
MidTool: Mid-training Data Synthesis for Agentic Tool Use
In the paper 'MidTool: Mid-training Data Synthesis for Agentic Tool Use', researchers present a new open corpus construction pipeline called MidTool. This pipeline is designed to teach large language models how to use…
Workshop Summary Report for Cyber AI Profile Hybrid Workshop #2
NIST summarizes stakeholder discussion on a Cyber AI Profile and related controls for securing AI systems.
On the Applicability of Safety Nets: A Safety-By-Design Solution for Certifying Neural Networks
A research study presents the first comprehensive analysis of Safety Nets, a combination of neural network compression and lookup tables, as a Safety-by-Design approach for certifying AI systems in aviation. The study…
Open-Vocabulary 3D Object Detection with Co-Distillation Discovery and Dual Guidance Robust Training
The authors propose a new framework for open-vocabulary 3D object detection that addresses the limitations of existing methods by improving both the reliability of novel object discovery and the robustness of model…
Pandora's AI Model Routing Box: Efficient Allocation with Costly Value Estimation
Researchers have developed a system that routes AI queries to the most effective and efficient model based on costly value estimation. The system, called Pandora's Router, uses a trade-off between fast but noisy…
ReguSim: Evaluating LLM Agent Rule Grounding in Financial Compliance
Researchers have introduced ReguSim, a controlled financial-compliance environment, and ReguBench, a monitoring benchmark, to evaluate the performance of LLM agents in financial markets. The study finds that while…
ReguSim: Evaluating LLM Agent Rule Grounding in Financial Compliance
The paper introduces ReguSim, a controlled environment for evaluating Language Learning Models (LLM) in financial compliance. It presents ReguBench, a monitoring benchmark, to analyze four artifacts: reasoning…
Repo0: Design-Driven Zero-to-All Code Generation
The paper introduces Repo0, a framework for zero-to-all code generation that maintains an explicit architectural state using a Dual-Directed Acyclic Graph (Dual-DAG). It iteratively evolves component boundaries guided…
SAPO: Single-Rollout Autoregressive Policy Optimization for Agentic Reinforcement Learning
A new reinforcement learning method called Single-rollout Autoregressive Policy Optimization (SAPO) is proposed to address the memory overhead, computation efficiency, and performance issues of existing methods. SAPO…
Specification-delta-driven data governance: an empirical study of the «spec-delta» as the unit of change in lakehouse data platforms
The paper discusses the concept of Spec-Delta, which is the idea that changes in data platforms should be reviewed as units of change in requirements rather than just code changes. The paper proposes a formalized…
The 2026 AI Index Report
Stanford HAI's 2026 AI Index organizes measurement across research and development, technical performance, responsible AI, the economy, science, medicine, education, policy, and public opinion.
Task-CoEvolve: Efficient Harness Optimization via Adaptive Validation Task Selection
The paper presents Task-CoEvolve, a new approach to optimize language model agent harnesses efficiently by adaptively selecting validation tasks. It addresses the challenge of using fixed validation sets by co-evolving…
Task-CoEvolve: Efficient Harness Optimization via Adaptive Validation Task Selection
Researchers have developed a method called Task-CoEvolve to improve the efficiency of harness optimization for large language models (LLMs). This method co-evolves validation tasks with the harness by focusing on tasks…
The Third Restructuring of Software Form: From the Three-Tier Architecture to Storage, Models, and Agents
The paper suggests a third shift in software development, called Software 3.0, which focuses on context and reasoning determining behavior. The proposed terminal form includes a generalized database, a large model, and…
The Third Restructuring of Software Form: From the Three-Tier Architecture to Storage, Models, and Agents
The paper suggests a third paradigm shift in software form, moving from data determining behavior (Software 2.0) to context and reasoning determining behavior (Software 3.0). This new form is proposed to converge into…
Write Once, Run Everywhere: The Axon DSL for Shape-Safe and Framework-Agnostic LLM Architectures
Researchers introduce Axon, a strongly typed domain-specific language with Haskell-like syntax that aims to enable a write-once, run everywhere paradigm for Language Model architectures. This allows for efficient model…
A Theory of Post-hoc Debate Judgement
In this research paper published in 2026, Xiang Yin et al. propose a theory of post-hoc debate judgement for agentic AI. The authors discuss the use of debates to improve AI performance and user engagement, and how the…
A Theory of Post-hoc Debate Judgement
In this research, Xiang Yin and colleagues propose a theory of post-hoc judgement for debates used by agentic AI. The goal is to improve the performance, explainability, and user engagement of AI agents through debate…
Adaptive Memory and Reflection Multi-Agent System for Medical Question Answering
Researchers have developed an adaptive memory and reflection multi-agent system for medical question answering. This system, which uses specialized agents with dedicated memory and feedback, aims to improve the…
Adaptive Memory and Reflection Multi-Agent System for Medical Question Answering
Researchers introduce a new multi-agent system called Adaptive Memory and Reflection (AMR) for medical question answering. This system aims to address the limitations of existing systems by using specialized agents…
ADEPT: Accelerating Dexterity via Pre-Training and Post-Training using Reinforcement Learning
The paper introduces ADEPT, a reinforcement learning framework for training multi-fingered robots to perform complex tasks directly from raw visual and tactile perception. The framework involves pre-training a…
Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning
Researchers introduce Group-Calibrated On-Policy Distillation (GC-OPD), a method for long-context reasoning tasks that aims to address the mismatch between on-policy distillation and task-specific verifiers. The new…
Beyond the Transcript: Detecting Covert Co ordination in Latent Multi-Agent Communication
Researchers have developed Verifiable Latent Alignments (VLA), a framework for monitoring and controlling communication between language-model agents that occurs through hidden states not visible in public transcripts…
Beyond the Transcript: Detecting Covert Co ordination in Latent Multi-Agent Communication
Researchers have developed Verifiable Latent Alignments (VLA), a framework to monitor and steer private communication channels used by language model agents in multi-agent systems. The framework links the private…
Breaking the weakest link to evade vision language models
The authors investigate the vulnerability of Vision Language Models (VLMs) to adversarial attacks, which can disrupt or force the models' interpretation of visual inputs. They propose a gradient-based attack method…
Decomposing Wrong-Consensus Agreement in LLM Self-Consistency: A GPT-4.1 Case Study
The study presents an analysis of the inconsistency in the answers provided by Large Language Models (LLMs) when using majority voting to improve accuracy. The authors propose a quantitative measure called Gamma, which…
DeepWeaver: Bridging the Evidence Synthesis Gap in Open-Ended Question Answering
The paper introduces DeepWeaver, a framework designed to address the gap between evidence retrieval and generation in open-ended question answering systems. DeepWeaver organizes and synthesizes noisy and fragmented…
DentAgent: Evidence-Centric Multi-Agent Coordination for Multimodal Dental Reasoning
Researchers have developed DentAgent, an evidence-centric multi-agent framework to address the limitations of existing dental AI systems. This system integrates various modalities like domain knowledge, radiographs…
DentAgent: Evidence-Centric Multi-Agent Coordination for Multimodal Dental Reasoning
A new system called DentAgent is introduced that aims to improve the accuracy and reliability of dental assessments by integrating various forms of evidence from different modalities. The system uses an…
Do Large Language Models Hallucinate Electric Fata Morganas?
The paper explores the issue of AI hallucinations (outputs that are made up or contradict source material) and their implications for machine consciousness. The author investigates causes of hallucinations in large…
Epistemic Subordination: Generative AI and the Infrastructure of Knowledge
The paper discusses how Generative AI encodes the dominant culture's way of knowing into its architecture, creating an epistemic condition known as 'epistemic subordination'. This affects minority epistemologies and…
Eureka: Task-Conditioned Meta-Agent Orchestration for Scientific Discovery
The paper presents Eureka, a Meta-Agent architecture that can complete long-horizon tasks by forming specialized Macro-Agents for specific tasks. The architecture allows for cost-benefit-gated evolution to update the…
Eureka: Task-Conditioned Meta-Agent Orchestration for Scientific Discovery
The paper introduces Eureka, a Meta-Agent architecture that can complete long-horizon tasks by forming specialized Macro-Agents through receding-horizon planning and architecture promotion. The agents can evolve to…
Eureka: Task-Conditioned Meta-Agent Orchestration for Scientific Discovery
The paper introduces Eureka, a meta-agent system designed for scientific discovery tasks. Eureka forms specialized agents through planning, promotion, and minimal-sufficient compilation, adapting to recurring…
Finetuning Strategies for Querying Sounds by Vocal Imitation
The authors describe their winning strategy for a sound effect querying competition in 2025, which involved two fine-tuning strategies: contrastive learning with a frozen pretrained CED encoder and joint…
From Threat Intelligence to Detection: Knowledge-driven Enrichment and Template-based Rule Grounding for Automated Sigma Rule Generation
The paper presents AUTOSIGMA, a system that automates the transformation of unstructured cyber threat intelligence (CTI) reports into relevant Sigma rules for threat detection. Instead of solely relying on language…
Graphical Design of Interpretable Architectures
The text discusses a new graphical notation for designing interpretable AI architectures, which addresses the limitations of existing representations. The proposed notation provides a global view of an architecture and…
Grouping the Stochastic Machine: Precision, Not Capability, as the Frontier Metric for AI Systems
The author argues that precision, rather than capability, should be the focus when comparing and measuring AI systems. Precision refers to how consistently an AI system performs the same task, while capability is where…
Harness Continual Learning: Continual Adaptation Beyond Model Parameters
The paper presents a new paradigm for continual learning called Harness Continual Learning (HCL). In this approach, an agent can adapt through a harness of prompts, memories, tools, skills, and routing rules. The…
Harness Continual Learning: Continual Adaptation Beyond Model Parameters
The paper 'Harness Continual Learning: Continual Adaptation Beyond Model Parameters' introduces a new approach to continual learning, focusing on the evolution of an agent's harness (which includes prompts, memories…
Identifying Implicit Premises for Logical Reconstruction of Argument Graphs
The authors propose a neuro-symbolic pipeline to generate implicit premises from natural language text, which can be translated into logical formulae to show the logical relationships between statements (entailment…
Interpretable AI predicts a 2026 summer dry anomaly in central China
Researchers have developed an AI model that predicts a dry anomaly in central China during the summer of 2026. This prediction is based on atmospheric circulation patterns, particularly a persistent cyclonic…
Metrics That Write Themselves: Evolving an Evaluator from Its Own Blind Spots
Researchers developed a method called EvalCEGAR to automatically create an evaluation metric for AI agents in tasks like report generation, where traditional metrics are difficult to develop. The method involves…
MLREF: Efficient Module Reuse for Reward Design in Reinforcement Learning via Large Language Models
The research proposes a new framework called Module Level Reward Evolution Framework (MLREF) to address the issue of unreliable reward function preservation and reuse in reinforcement learning. MLREF uses a module…
ORBITER: Conflict-Aware Decision-Making for Agentic Last-Mile Delivery
The paper introduces ORBITER, a system designed to make decisions for last-mile delivery by couriers. It models the courier service through decision points and uses language models (LLMs) to reason about individual…
PGFS++: Molecular Property Improvement under Synthesis and Diversity Constraints
Researchers present PGFS++, an advancement of a previous method called PGFS for improving molecular properties like drug-likeness or binding affinity. The new method addresses limitations in PGFS by using trainable…
Pre-Compiled Pipeline Shards for Distributed LLM Inference on Intel AI PC Fleets
The authors present a method to split large machine learning models into smaller parts called pipeline shards, which can then be run on multiple Integrated AI Processors (AIPCs) over an ordinary network. They show that…
Page 5 of 47
