HN Hd1633 5/14/2026

How LLM Inference Works

Technical explanation of LLM inference mechanics covering tokenization, transformer architecture, and token-by-token generation.

HN satoshiclad 5/14/2026

Multi-LLM AI trading agent harness

Trading agent harness (1rok) comparing multiple LLM models (OpenAI, Anthropic, Gemini, DeepSeek) against financial tools with live performance leaderboard.

HN supercrafthost 5/14/2026

Vector embeddings are the wrong default for AI agent memory

Analysis of vector database limitations for agent memory with production failure modes; proposes structured KV and MCP memory servers as alternatives.

HN ChicknNuggt 5/14/2026

A Streaming First Language for Generative UI

OpenUI is a streaming-first Generative UI framework with React runtime and component libraries, 67% more token-efficient than JSON for LLM applications.

HN baruch 5/14/2026

Storage based KVCache for denser token factory

Title-only submission on storage-based KVCache optimization for token generation. Insufficient content to evaluate properly.

HN amitbidlan 5/14/2026

Tracing and tenant-isolation firewall for AI agents (Apache 2.0)

Open-source operational platform for LLM agents in production. Self-hosted Docker, tracing, governance policies, OWASP guardrails, audit and alerts. Apache 2.0.

Ax Nishad Singhi, Christian Bialas, Snehal Jauhri, Vignesh Prasad, Georgia Chalvatzaki, Marcus Rohrbach, Anna Rohrbach 5/14/2026

Think Twice, Act Once: Verifier-Guided Action Selection For Embodied Agents

Verifier-Guided Action Selection improves embodied agents using multimodal LLMs with verification step to handle out-of-distribution scenarios.

Ax Wo Wei Lin, Ethan Rathbun, Enrico Marchesini Xiang Zhi Tan 5/14/2026

Macro-Action Based Multi-Agent Instruction Following through Value Cancellation

Multi-agent reinforcement learning approach using value cancellation to handle conflicting natural language instructions interrupting macro-actions.

Ax Hao Wang, Hanchen Li, Qiuyang Mang, Alvin Cheung, Koushik Sen, Dawn Song 5/14/2026

Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack

BenchJack framework and taxonomy of eight flaw patterns for auditing agent benchmarks to prevent reward hacking and ensure benchmarks are secure by design.

Ax Alina Hyk, Sandhya Saisubramanian 5/14/2026

Learning Transferable Latent User Preferences for Human-Aligned Decision Making

Method for learning transferable latent user preferences to improve human-aligned decision-making in LLM reasoning modules without extensive repeated feedback.

Ax Eugenia Kim, Ioana Tanase, Christina Mallon 5/14/2026

DisaBench: A Participatory Evaluation Framework for Disability Harms in Language Models

DisaBench taxonomy and evaluation framework for assessing disability-related harms in LLMs, co-created with disability advocates and red team experts.

Ax Tommaso Giovannelli, Griffin D. Kent 5/14/2026

CHAL: Council of Hierarchical Agentic Language

CHAL proposes hierarchical council structure for multi-agent LLM debate to improve reasoning beyond majority voting through dialectic systems.

Ax Sungheon Jeong, Ryozo Masukawa, Sanggeon Yun, Mahdi Imani, Mohsen Imani 5/14/2026

State-Centric Decision Process

State-Centric Decision Process runtime framework enables LLM agents to operate in text-based environments by constructing explicit state spaces and transitions.

Ax Sridhar Mahadevan 5/14/2026

PROMETHEUS: Automating Deep Causal Research Integrating Text, Data and Models

PROMETHEUS framework combines LLM-extracted causal claims from text, data, code, and simulations into navigable causal world models and predictive-state models.

Ax Anamika Ragu, Aneesh Jonelagadda 5/14/2026

Multimodal Hidden Markov Models for Persistent Emotional State Tracking

Lightweight framework using multimodal hidden Markov models to track emotional state across conversation turns for clinical and communication guidance.

Ax Ali Al-Lawati, Nafis Tripto, Abolfazl Ansari, Jason Lucas, Suhang Wang, Dongwon Lee 5/14/2026

Moltbook Moderation: Uncovering Hidden Intent Through Multi-Turn Dialogue

Multi-turn dialogue moderation system for detecting malicious intent in multi-agent systems through interaction patterns beyond surface content filtering.

Ax Harshita Chopra, Kshitish Ghate, Aylin Caliskan, Tadayoshi Kohno, Chirag Shah, Natasha Jaques 5/14/2026

Beyond Cooperative Simulators: Generating Realistic User Personas for Robust Evaluation of LLM Agents

Framework for generating diverse, realistic LLM-based user personas to evaluate robustness of LLM agents in handling unclear, impatient, or reluctant users.

Ax Vardhan Dongre, Joseph Hsieh, Viet Dac Lai, Seunghyun Yoon, Trung Bui, Dilek Hakkani-T\"ur 5/14/2026

When Attention Closes: How LLMs Lose the Thread in Multi-Turn Interaction

Mechanistic explanation of LLM instruction degradation in long multi-turn interactions via channel transitions in attention and residual streams.

Ax James M. Mazzu 5/14/2026

Sustaining AI safety: Control-theoretic external impossibility, intrinsic necessity, and structural requirements

Control-theoretic analysis of whether externally-enforced AI safety strategies can sustain safety as external control becomes unreliable.

Ax Junwei Liao, Shuai Li, Muning Wen, Jun Wang, Weinan Zhang 5/14/2026

Position: Agentic AI System Is a Foreseeable Pathway to AGI

Position paper arguing Agentic AI systems are necessary for AGI over monolithic scaling, contrasting optimization constraints theoretically.

Ax Jiashuo Sun, Jimeng Shi, Yixuan Xie, Saizhuo Wang, Jash Rajesh Parekh, Pengcheng Jiang, Zhiyi Shi, Jiajun Fan, Qinglong Zheng, Peiran Li, Shaowen Wang, Ge Liu, Jiawei Han 5/14/2026

Retrieval is Cheap, Show Me the Code: Executable Multi-Hop Reasoning for Retrieval-Augmented Generation

Executable code-based multi-hop reasoning for RAG with explicit intermediate states addressing drift and error propagation in multi-step retrieval.

Ax Dylan Zhang, Yanshan Lin, Zhengkun Wu, Yihang Sun, Bingxuan Li, Dianqi Li, Hao Peng 5/14/2026

Useful Memories Become Faulty When Continuously Updated by LLMs

Studies degradation of continuous LLM-updated memory banks in agentic systems, showing consolidated abstractions become faulty with frequent updates.

Ax Mragisha Jain, Tirth Bhatt, Griffin Pitts, Aum Pandya, Peter Brusilovsky, Narges Norouzi, Arto Hellas, Juho Leinonen, Bita Akram 5/14/2026

Retrieval-Augmented Tutoring for Algorithm Tracing and Problem-Solving in AI Education

KITE: RAG-based intelligent tutoring system using intent-aware Socratic responses for algorithm tracing and problem-solving education.

Ax Yuxin Liu, Ziang Ye, Yueqing Sun, Mingye Zhu, Jinwei Xiao, Zhuowen Han, Qi GU, Xunliang Cai, Lei Zhang 5/14/2026

MAP: A Map-then-Act Paradigm for Long-Horizon Interactive Agent Reasoning

MAP paradigm for interactive LLM agents that establishes environmental understanding upfront via mapping, addressing delayed perception and epistemic bottlenecks.

Ax Giuliano Lorenzoni, Paulo Alencar, Donald Cowan 5/14/2026

An Agentic LLM-Based Framework for Population-Scale Mental Health Screening

Agentic LLM framework for population-scale mental health screening processing clinical data from electronic records and telemedicine platforms.

Ax Junjie Li, Ziao Wang, NingXuan Ma, Jianghong Ma, Xiaofeng Zhang 5/14/2026

GRACE: Gradient-aligned Reasoning Data Curation for Efficient Post-training

GRACE: gradient-aligned reasoning data curation that scores individual steps in traces for efficient LLM post-training, not whole samples.

Ax Rikui Huang, Shengzhe Zhang, Wei Wei 5/14/2026

Strikingness-Aware Evaluation for Temporal Knowledge Graph Reasoning

Strikingness-aware evaluation framework for temporal knowledge graph reasoning that emphasizes rare events requiring deeper reasoning over trivial repetitions.

Ax Moritz Firsching, Paul Lezeau, Salvatore Mercuri, Mikl\'os Z. Horv\'ath, Ya\"el Dillies, Calle S\"onne, Eric Wieser, Fred Zhang, Thomas Hubert, Blaise Ag\"uera y Arcas, Pushmeet Kohli 5/14/2026

Formal Conjectures: An Open and Evolving Benchmark for Verified Discovery in Mathematics

Formal Conjectures benchmark: 2615 mathematical problems in Lean 4 including 1029 open research conjectures for evaluating automated reasoning systems.

Ax Hao Zhou, Tiru Wu, Yan Jiang, Wanqi Zhou, Junxing Hu, Ai Han 5/14/2026

Hierarchical Attacks for Multi-Modal Multi-Agent Reasoning

Studies adversarial attack vulnerabilities in multi-modal multi-agent systems, extending prior work beyond isolated or unimodal settings.

Ax Hanwen Zhang, Dusit Niyato, Wei Zhang, Xin Lou, Malcolm Yoke Hean Low 5/14/2026

An Agentic AI Framework with Large Language Models and Chain-of-Thought for UAV-Assisted Logistics Scheduling with Mobile Edge Computing

Agentic AI framework using LLMs and chain-of-thought for UAV logistics scheduling and mobile edge computing task allocation.

Ax Yuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang, Wei Hu 5/14/2026

Improving Code Translation with Syntax-Guided and Semantic-aware Preference Optimization

Syntax-guided and semantic-aware preference optimization for improving LLM code translation with robust rewards derived from source code.

Ax Marios Adamidis, Danae Katrisioti, Yannis Tzitzikas, Emmanuel Stratakis 5/14/2026

It's not the Language Model, it's the Tool: Deterministic Mediation for Scientific Workflows

Proposes typed mediation pattern where LLMs orchestrate deterministic tools for reproducible scientific workflows instead of generating analyses directly.

Ax Junlong Ke, Zichen Wen, Weijia Li, Conghui He, Linfeng Zhang 5/14/2026

Respecting Self-Uncertainty in On-Policy Self-Distillation for Efficient LLM Reasoning

EGRSD method for on-policy self-distillation that weights token-level supervision by teacher prediction entropy to improve LLM reasoning efficiency.

Ax Yucheng Guo, Yongjian Guo, Zhong Guan, Wen Huang, Haoran Sun, Haodong Yue, Xiaolong Xiang, Shuai Di, Zhen Sun, Luqiao Wang, Junwu Xiong, Yicheng Gong 5/14/2026

D-VLA: A High-Concurrency Distributed Asynchronous Reinforcement Learning Framework for Vision-Language-Action Models

D-VLA framework for distributed asynchronous RL training of Vision-Language-Action models addressing resource bottlenecks in embodied AI.

Ax Jonas Reiter, Jakob Elias Gebler, Hector Geffner 5/14/2026

Differentiable Learning of Lifted Action Schemas for Classical Planning

Presents differentiable learning method for lifted action schemas in classical planning to enable structural generalization across domain instances.

Ax Miko{\l}aj Langner, Dzmitry Pihulski, Jan Eliasz, Micha{\l} Rajkowski, Przemys{\l}aw Kazienko, Maciej Piasecki, Jan Koco\'n, Teddy Ferdinan 5/14/2026

What properties of reasoning supervision are associated with improved downstream model quality?

Proposes quantitative metrics to predict reasoning dataset utility before training, evaluating on 8B-11B models with Polish reasoning dataset variants.

Ax Yuanzhe Wang, Tian Zhi, Zihang Wei, Hongguang Wang, Jiaming Guo, Yang Zhao, Zisheng Liu, Shiyu Quan, Xing Hu, Zidong Du, Yunji Chen 5/14/2026

Discrete Diffusion for Complex and Congested Multi-Agent Path Finding with Sparse Social Attention

arXiv paper on discrete diffusion for multi-agent path finding with sparse social attention in dense, congested environments. Coordination complexity research.

Ax Yafu Li, Runzhe Zhan, Haoran Zhang, Shunkai Zhang, Yizhuo Li, Zhilin Wang, Jiacheng Chen, Futing Wang, Xuyang Hu, Yuchen Fan, Bangjie Xu, Yucheng Su, Xinmiao Han, Chenxi Li, Haodi Lei, Yufeng Zhao, Zejin Lin, Qianjia Cheng, Tong Zhu, Xiaoye Qu, Ganqu Cui, Peng Ye, Yun Luo, Zhouchen Lin, Yu Qiao, Bowen Zhou, Ning Ding, Yu Cheng 5/14/2026

Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling

arXiv paper on scaling recipe converting reasoning models into olympiad-level solvers achieving gold-medal performance on IMO/IPhO problems.

Ax Joy Bose 5/14/2026

IdeaForge: A Knowledge Graph-Grounded Multi-Agent Framework for Cross-Methodology Innovation Analysis and Patent Claim Generation

arXiv paper on IdeaForge, knowledge graph-grounded multi-agent framework applying multiple innovation methodologies (TRIZ, Design Thinking) for patent claim generation.

Ax Luca Belli, Kate H. Bentley, Josh Gieringer, Emily Van Ark, Nilu Zhao, Pradip Thachile, Matt Hawrilenko, Millard Brown, Adam M. Chekroud 5/14/2026

VERA-MH: Validation of Ethical and Responsible AI in Mental Health

arXiv paper introducing VERA-MH, clinically-validated evaluation framework for chatbot safety in mental health contexts, focusing on suicidal ideation response assessment.

Ax Qinchuan Cheng, Zhantao Gong, Pengzhan Sun, Angela Yao, Xulei Yang, Shijie Li 5/14/2026

Ego2World: Compiling Egocentric Cooking Videos into Executable Worlds for Belief-State Planning

arXiv paper compiling egocentric cooking videos into executable simulations for embodied agents to perform belief-state planning under partial observation.

Ax Markus Wenzel, Tobias Strapatsas, Jessika Kress, Dorothea Sauer, Nele Gessler, Horst K. Hahn 5/14/2026

Multi-Agent Systems in Emergency Departments: Validation Study on a ED Digital Twin

arXiv paper validating multi-agent systems in emergency department simulation using hybrid discrete event and agent-based modeling.

Ax Liangtian Liu, Zeyuan Wang, Ziyu Li, Kai Ouyang, Zichao Tang, Chengfu Liu, Haifeng Li, Hanwen Yu, Wentao Yang, Cheng Yang, Dongyang Hou 5/14/2026

RS-Claw: Progressive Active Tool Exploration via Hierarchical Skill Trees for Remote Sensing Agents

arXiv paper on RS-Claw framework enabling remote sensing agents to autonomously explore and invoke image-processing tools via hierarchical skill trees.

Ax Zabir Al Nazi, Shubhashis Roy Dipta 5/14/2026

TRIAGE: Evaluating Prospective Metacognitive Control in LLMs under Resource Constraints

arXiv paper on prospective metacognitive control in LLMs: agents deciding which tasks to attempt and compute allocation under token budget constraints.

Ax Suli Wang, Yiqun Duan, Yu Deng, Rundong Zhao, Dai Shi, Xinliang Zhou 5/14/2026

Cognifold: Always-On Proactive Memory via Cognitive Folding

arXiv paper introducing Cognifold, brain-inspired proactive agent memory that continuously organizes experiences into cognitive structures. Agent architecture research.

Ax Samuel Schapiro, Alexi Gladstone, Jonah Black, Heng Ji 5/14/2026

Assessing the Creativity of Large Language Models: Testing, Limits, and New Frontiers

arXiv paper on measuring LLM creativity using automated tests adapted from human creativity assessments. Research on evaluation methodology.

Ax Kangning Zhang, Shuai Shao, Qingyao Li, Jianghao Lin, Lingyue Fu, Shijian Wang, Wenxiang Jiao, Yuan Lu, Weiwen Liu, Weinan Zhang, Yong Yu 5/14/2026

MMSkills: Towards Multimodal Skills for General Visual Agents

MMSkills framework for building reusable multimodal skills for visual agents that encode procedural knowledge beyond text/code through visual state recognition.

Ax Juho Leinonen, Lisa Zhang, Arto Hellas 5/14/2026

AI-Generated Slides: Are They Good? Can Students Tell?

Study evaluating GenAI tools (NotebookLM, Claude, Copilot, Cursor) for generating educational slides from course notes, assessing instructor and student perceptions.

Ax Jiabei Liu, Wenyu Mao, Junfei Tan, Chunxu Shen, Lingling Yi, Jiancan Wu, Xiang Wang 5/14/2026

Scaling Retrieval-Augmented Reasoning with Parallel Search and Explicit Merging

MultiSearch method scaling retrieval-augmented reasoning with parallel search queries and explicit merging to improve signal-to-noise ratios in LLM reasoning.

Ax Chengzhi Shen (Cherise), Weixiang Shen (Cherise), Tobias Susetzky (Cherise), Chen (Cherise), Chen (Cherise), Jun Li, Yuyuan Liu, Xuepeng Zhang, Zhenyu Gong, Daniel Rueckert, Jiazhen Pan 5/14/2026

RealICU: Do LLM Agents Understand Long-Context ICU Data? A Benchmark Beyond Behavior Imitation

RealICU benchmark testing whether LLM agents understand long-context ICU clinical data beyond behavior imitation for medical decision support.