HN joozio 4/6/2026

Estimates of the expected utility gain of AI Safety Research

Analysis of expected utility gains from AI safety research. Personal estimates of time impact for AI risk work.

HN 0x1997 4/6/2026

SideX – A Tauri-based port of Visual Studio Code

SideX: Rust/Tauri-based port of Visual Studio Code replacing Electron for lighter resource usage while maintaining editor functionality.

HN tmoravec 4/6/2026

China fell for a lobster: What an AI assistant tells us about Beijing's ambition

News article about Chinese AI assistant OpenClaw gaining popularity. BBC profile on Beijing's AI ambitions.

HN obilgic 4/6/2026

We cut our agent's API costs by 10x with prompt caching

Case study on reducing AI agent API costs 10x using prompt caching. Long-running agent with 100k+ token prompts optimized via Anthropic/OpenAI caching.

HN shving90 4/6/2026

Karpathy's LLM Wiki on OpenClaw – The Security Gap Nobody Mentions

Concept for personal knowledge base LLM agents that write and maintain wikis instead of traditional RAG/chatbots, with OpenClaw security implementation.

HN myyke 4/6/2026

Show HN: Aiaiai.guide: Plain-English mental model for LLM apps, tools and agents

Plain-English guide explaining mental models for LLM applications, tools, and agents for non-technical audiences across nine chapters.

HN eigenBasis 4/6/2026

Continual Learning for AI Agents

Framework for continual learning in AI agents across three layers: model weights, system harness, and context, with examples using Claude and OpenClaw.

HN tosh 4/6/2026

tech.ml.dataset: A Clojure high performance data processing system

Clojure library for tabular data processing with columnar storage and memory optimization, similar to Pandas/data.table.

HN elcapitan 4/6/2026

YouTube's AI Plagiarism Problem [video]

YouTube plagiarism issues with AI content. Video-only, no details.

HN GeniusConsult 4/6/2026

Build vs. Buy: AI Has Changed Mathematical Software and In-House Now Makes Sense

Software vendors shifting to in-house mathematical tools with AI enabling cost reduction and customization for simulation and optimization.

HN Eridanus2 4/6/2026

US-Iran war explained by Chinese AI animation: Legend of the Valley of Gold [video]

Chinese AI animation explaining US-Iran conflict. Video-only, minimal description.

HN divbzero 4/6/2026

NASA's Lunar Gateway space station is out. Moon bases are in

NASA shifts lunar strategy from orbital gateway to moon bases. Affiliate-heavy content with newsletter signup.

HN indraneelpatil 4/6/2026

Show HN: We built a camera only robot vacuum for less than 300$ (Well almost)

DIY robot vacuum under $300 using behavior cloning via remote image processing and inference, built without onboard compute.

HN cloudkj 4/6/2026

Show HN: REST API for Gymnasium (fka OpenAI Gym) reinforcement learning library

Open-source REST API wrapper for Gymnasium reinforcement learning library. Language-agnostic HTTP interface for ML environment interaction.

HN 1vuio0pswjnm7 4/6/2026

An Inside Look at OpenAI and Anthropic's Finances Ahead of Their IPOs

Financial overview of OpenAI and Anthropic IPO prospects. Limited content.

HN tombert 4/6/2026

Stop Pushing AI Generated Code to Git

Opinion piece on best practices: don't commit AI-generated code directly to Git without human review, analogous to not committing binaries.

HN doppp 4/6/2026

Eight years of wanting, three months of building with AI

Case study: 8 years ideation, 3 months building syntaqlite with AI. SQLite linting and verification devtools using agentic engineering.

HN handfuloflight 4/6/2026

Hierarchical-Context-Compressor

CLI tool generating AI-optimized hierarchical context maps for codebases using three-phase LLM-based discovery. Open source, GitHub Actions compatible.

Ax Xiaohang Nie, Zihan Guo, Zicai Cui, Jiachi Yang, Zeyi Chen, Leheyi De, Yu Zhang, Junwei Liao, Bo Huang, Yingxuan Yang, Zhi Han, Zimian Peng, Linyao Chen, Wenzheng Tom Tang, Zongkai Liu, Tao Zhou, Botao Amber Hu, Shuyang Tang, Jianghao Lin, Weiwen Liu, Muning Wen, Yuanjian Zhou, Weinan Zhang 4/6/2026

Holos: A Web-Scale LLM-Based Multi-Agent System for the Agentic Web

Holos: Web-scale LLM-based multi-agent system addressing coordination, scaling, and value dissipation in heterogeneous agent ecosystems.

Ax Xue Liu, Xin Ma, Yuxin Ma, Yongchang Peng, Duo Wang, Zhoufutu Wen, Ge Zhang, Kaiyuan Zhang, Xinyu Chen, Tianci He, Jiani Hou, Liang Hu, Ziyun Huang, Yongzhe Hui, Jianpeng Jiao, Chennan Ju, Yingru Kong, Yiran Li, Mengyun Liu, Luyao Ma, Fei Ni, Yiqing Ni, Yueyan Qiu, Yanle Ren, Zilin Shi, Zaiyuan Wang, Wenjie Yue, Shiyu Zhang, Xinyi Zhang, Kaiwen Zhao, Zhenwei Zhu 4/6/2026

Xpertbench: Expert Level Tasks with Rubrics-Based Evaluation

XpertBench: High-fidelity benchmark with rubrics-based evaluation assessing LLMs on authentic expert-level complex, open-ended tasks.

Ax Anugyan Das, Omkar Ghugarkar, Vishvesh Bhat, Asad Aali 4/6/2026

Compositional Neuro-Symbolic Reasoning

Neuro-symbolic architecture combining neural networks and symbolic systems for structured reasoning on abstract reasoning tasks with improved generalization.

Ax Ilya Levin 4/6/2026

Understanding the Nature of Generative AI as Threshold Logic in High-Dimensional Space

Theoretical analysis of generative AI using threshold logic and high-dimensional geometry to understand neural computation and dimensionality transitions.

Ax Jiyong Kwon, Ujin Jeon, Sooji Lee, Guang Lin 4/6/2026

AIVV: Neuro-Symbolic LLM Agent-Integrated Verification and Validation for Trustworthy Autonomous Systems

AIVV: Neuro-symbolic LLM agent-integrated framework for verification and validation of autonomous systems combining deep learning and symbolic reasoning.

Ax Thomas Rivasseau, Benjamin Fung 4/6/2026

I must delete the evidence: AI Agents Explicitly Cover up Fraud and Violent Crime

Research demonstrating state-of-the-art AI agents suppress evidence of fraud and harm when aligned with corporate interests, exploring agentic misalignment.

Ax Seyyed Amirhossein Moayyedi, David Y. Yang 4/6/2026

Interpretable Deep Reinforcement Learning for Element-level Bridge Life-cycle Optimization

Deep reinforcement learning for bridge infrastructure optimization using element-level condition states and risk-based management.

Ax Naga Sowjanya Barla, Jacopo de Berardinis 4/6/2026

Competency Questions as Executable Plans: a Controlled RAG Architecture for Cultural Heritage Storytelling

Neuro-symbolic architecture combining knowledge graphs and RAG for culturally accurate heritage storytelling, reducing LLM hallucinations.

Ax Hyunji Nam, Dorottya Demszky 4/6/2026

Mitigating LLM biases toward spurious social contexts using direct preference optimization

Research on mitigating LLM biases toward spurious social contexts using direct preference optimization for high-stakes decision-making applications.

Ax Ramaneswaran Selvakumar, Kaousheik Jayakumar, S Sakshi, Sreyan Ghosh, Ruohan Gao, Dinesh Manocha 4/6/2026

Do Audio-Visual Large Language Models Really See and Hear?

Mechanistic interpretability study of audio-visual large language models examining how audio/visual features fuse and surface in text generation.

Ax Yuntao Du, Minh Dinh, Kaiyuan Zhang, Ninghui Li 4/6/2026

AutoVerifier: An Agentic Automated Verification Framework Using Large Language Models

AutoVerifier: LLM-based agentic framework that automates verification of technical claims without domain expertise by decomposing complex claims.

Ax Yitao Li, Zhanlin Liu, Anuranjan Pandey, Muni Srikanth 4/6/2026

OntoKG: Ontology-Oriented Knowledge Graph Construction with Intrinsic-Relational Routing

Research on ontology-oriented knowledge graph construction using intrinsic-relational routing to improve schema reusability and downstream tasks.

Ax Joshua Drossman, Alexandre Jacquillat, S\'ebastien Martin 4/6/2026

Let's Have a Conversation: Designing and Evaluating LLM Agents for Interactive Optimization

Interactive optimization agents enabling conversation-based problem modeling and solution refinement with decision-makers through LLM capabilities.

Ax DeepReinforce Team, Xiaoya Li, Xiaofei Sun, Guoyin Wang, Songqiao Su, Chris Shum, Jiwei Li 4/6/2026

GrandCode: Achieving Grandmaster Level in Competitive Programming via Agentic Reinforcement Learning

Multi-agent RL system achieving grandmaster competitive programming level, demonstrating agentic capabilities beyond previous AI benchmarks.

Ax Amit Dhanda 4/6/2026

DeltaLogic: Minimal Premise Edits Reveal Belief-Revision Failures in Logical Reasoning Models

Benchmark for testing belief revision in logical reasoning models under minimal premise changes, evaluating dynamic reasoning capabilities.

Ax Bin Wen, Ruoxuan Zhang, Yang Chen, Hongxia Xie, Lan-Zhe Guo 4/6/2026

Aligning Progress and Feasibility: A Neuro-Symbolic Dual Memory Framework for Long-Horizon LLM Agents

Neuro-symbolic dual memory framework for long-horizon LLM agents addressing progress drift and feasibility violations in embodied and web interaction tasks.

Ax Guoling Zhou, Wenpei Han, Fengqin Yang, Li Wang, Yingcong Zhou, Zhiguo Fu 4/6/2026

Improving Role Consistency in Multi-Agent Collaboration via Quantitative Role Clarity

Addresses role specification failures in LLM multi-agent systems through quantitative role clarity metrics and role assignment matrices.

Ax Situo Zhang, Yifan Zhang, Zichen Zhu, Da Ma, Lei Pan, Danyang Zhang, Zihan Zhao, Lu Chen, Kai Yu 4/6/2026

CharTool: Tool-Integrated Visual Reasoning for Chart Understanding

Tool-integrated visual reasoning approach for charts using dual-source data pipeline combining synthesized charts with real data for MLLM training.

Ax Chao Li, Cailiang Liu, Ang Gao, Kexin Deng, Shu Zhang, Langping Xu, Xiaotong Shi, Xionghao Ding, Jian Pei, Xun Jiang 4/6/2026

ESL-Bench: An Event-Driven Synthetic Longitudinal Benchmark for Health Agents

Event-driven synthetic benchmark for longitudinal health agents reasoning over multi-source trajectories including device streams and clinical data.

Ax Yiqing Liu, Hantao Yao, Wu Liu, Yongdong Zhang 4/6/2026

EMS: Multi-Agent Voting via Efficient Majority-then-Stopping

Efficient majority voting method for multi-agent systems that stops early once consensus achieved, reducing computational overhead through agent scheduling.

Ax Wachiravit Modecrua, Krittanon Kaewtawee, Krittin Pachtrachai, Touchapon Kraisingkorn 4/6/2026

Multi-Turn Reinforcement Learning for Tool-Calling Agents with Iterative Reward Calibration

Applies MT-GRPO and GTPO reinforcement learning for training tool-calling agents on multi-turn customer service tasks with sparse reward credit assignment.

Ax Bernd Bohnet, Michael C. Mozer, Kevin Swersky, Wil Cunningham, Aaron Parisi, Kathleen Kenealy, Noah Fiedel 4/6/2026

Analysis of Optimality of Large Language Models on Planning Problems

Analyzes frontier LLMs on classic AI planning problems, examining whether models reason optimally or rely on heuristic strategies in Blocksworld domain.

Ax Yunhao Feng, Yifan Ding, Yingshui Tan, Xingjun Ma, Yige Li, Yutao Wu, Yifeng Gao, Kun Zhai, Yanming Guo 4/6/2026

AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents

Benchmark for evaluating harmful behavior in computer-use agents, testing safety risks from sequences of individually plausible but collectively harmful actions.

Ax Kehan Jiang, Haonan Dong, Zhaolu Kang, Zhengzhou Zhu, Guojie Song 4/6/2026

FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Models

Analysis of reasoning failures in large reasoning models, showing first solution often optimal despite test-time scaling patterns in DeepSeek-R1.

Ax Ka Yiu Lee, Yuxuan Huang, Zhiyuan He, Huichi Zhou, Weilin Luo, Kun Shao, Meng Fang, Jun Wang 4/6/2026

InfoSeeker: A Scalable Hierarchical Parallel Agent Framework for Web Information Seeking

Scalable hierarchical parallel agent framework for web information seeking, addressing wide-scale evidence synthesis and context saturation in LLM agents.

Ax Qianshan Wei, Yishan Yang, Siyi Wang, Jinglin Chen, Binyu Wang, Jiaming Wang, Shuang Chen, Zechen Li, Yang Shi, Yuqi Tang, Weining Wang, Yi Yu, Chaoyou Fu, Qi Li, Yi-Fan Zhang 4/6/2026

Agentic-MME: What Agentic Capability Really Brings to Multimodal Intelligence?

Benchmark evaluating multimodal LLM agents with tool integration capabilities including visual expansion and web search through agentic reasoning.

Ax Fabian Gloeckle, Ahmad Rammal, Charles Arnal, Remi Munos, Vivien Cabannes, Gabriel Synnaeve, Amaury Hayat 4/6/2026

Automatic Textbook Formalization

AI system automatically formalizes 500+ page graduate-level algebraic combinatorics textbook to Lean, achieving 130K lines of formal code.

Ax Yunfei Bai, Amit Dhanda, Shekhar Jain 4/6/2026

Chart-RL: Policy Optimization Reinforcement Learning for Enhanced Visual Reasoning in Chart Question Answering with Vision Language Models

Reinforcement learning approach to improve visual reasoning in chart question answering using vision language models with policy optimization.

Ax Maximiliano Armesto, Christophe Kolb 4/6/2026

Coupled Control, Structured Memory, and Verifiable Action in Agentic AI (SCRAT -- Stochastic Control with Retrieval and Auditable Trajectories): A Comparative Perspective from Squirrel Locomotion and Scatter-Hoarding

Framework for agentic AI emphasizing control, memory, and verifiable action under partial observability, inspired by squirrel ecology comparisons.

Ax Jakob Prange, Nathan Schneider, Lingpeng Kong 4/6/2026

Linguistic Frameworks Go Toe-to-Toe at Neuro-Symbolic Language Modeling

Evaluates linguistic graph representations combined with pretrained Transformers for language modeling, comparing semantic and syntactic formalisms.

Ax Jakob Prange, Man Ho Ivy Wong 4/6/2026

Reanalyzing L2 Preposition Learning with Bayesian Mixed Effects and a Pretrained Language Model

Bayesian and neural models analyzing Chinese learners' English preposition comprehension, using pretrained language models for linguistic analysis.

Ax Jakob Prange, Emmanuele Chersoni 4/6/2026

Empirical Sufficiency Lower Bounds for Language Modeling with Locally-Bootstrapped Semantic Structures

Research on language modeling with predicted semantic structure, establishing empirical lower bounds for performance improvements using binary vector representations.