Ax Nina Corvelo Benz, Stratis Tsirtsis, Eleni Straitouri, Ivi Chatzi, Ander Artola Velasco, Suhas Thejaswi, Manuel Gomez-Rodriguez 3/26/2026

Evaluation of Large Language Models via Coupled Token Generation

Causal framework for evaluating LLMs controlling for randomization in token generation. Proposes coupled generation model for fair model comparison and ranking.

Ax Judy Hanwen Shen, Ellen Vitercik, Anders Wikum 3/26/2026

Algorithms with Calibrated Machine Learning Predictions

Framework integrating ML prediction uncertainty into online algorithm design. Uses calibration to leverage prediction-level confidence in algorithms with predictions.

Ax Andreas Panayiotou, Panayiotis Charalambous, Ioannis Karamouzas 3/26/2026

Gen-C: Populating Virtual Worlds with Generative Crowds

Gen-C: Generative framework for simulating high-level crowd behaviors in virtual environments. Captures agent-agent and agent-environment interactions over time.

Ax Shubham Kumar Nigam, Balaramamahanthi Deepak Patnaik, Noel Shallum, Kripabandhu Ghosh, Arnab Bhattacharya 3/26/2026

Structured Legal Document Generation in India: A Model-Agnostic Wrapper Approach with VidhikDastaavej

VidhikDastaavej: Model-agnostic wrapper for automated legal document generation in Indian context. Introduces large-scale anonymized dataset for long-form legal drafting.

Ax Rodrigo P\'erez Ortiz, Gibbs Nwemadji, Jean Barbier, Federica Gerace, Alessandro Ingrosso, Clarissa Lauditi, Enrico M. Malatesta 3/26/2026

Generalization performance of narrow one-hidden layer networks in the teacher-student setting

Theoretical analysis of generalization in one-hidden-layer neural networks using teacher-student framework. Provides complete characterization for generic activation functions.

Ax Zhihao Luo, Wentao Yan, Jingyu Gong, Min Wang, Zhizhong Zhang, Xuhong Wang, Yuan Xie, Xin Tan 3/26/2026

NaviMaster: Learning a Unified Policy for GUI and Embodied Navigation Tasks

Unified agent framework (NaviMaster) handling both GUI navigation and embodied navigation tasks via MDP formulation. First model to combine disparate domains with shared training paradigm.

Ax Yuan Wang, Mingyu Li, Haibo Chen 3/26/2026

From Imperative to Declarative: Towards LLM-friendly OS Interfaces for Boosted Computer-Use Agents

Declarative OS interfaces for computer-use agents to replace GUIs, enabling LLMs to execute high-level goals with fewer API calls and less decomposition.

Ax Yijie Xu, Huizai Yao, Zhiyu Guo, Pengteng Li, Aiwei Liu, Xuming Hu, Weiyu Guo, Hui Xiong 3/26/2026

You only need 4 extra tokens: Synergistic Test-time Adaptation for LLMs

SyTTA: Label-free test-time adaptation for LLMs in specialized domains using only 4 extra tokens to mitigate distribution shifts.

Ax Minjae Lee, Minsuk Kahng 3/26/2026

Data-Prompt Co-Evolution: Growing Test Sets to Refine LLM Behavior

Method for co-evolving test sets and prompts to refine LLM behavior, enabling iterative refinement of domain-specific policies without manual tuning.

Ax Arthur Jacot 3/26/2026

Deep Learning as a Convex Paradigm of Computation: Minimizing Circuit Size with ResNets

Theoretical analysis of deep neural networks as convex computation paradigm, examining how DNNs implement Occam's razor through circuit size minimization.

Ax Zhihao Zhan, Jiaying Zhou, Likui Zhang, Qinhan Lv, Hao Liu, Jusheng Zhang, Weizheng Li, Ziliang Chen, Tianshui Chen, Ruifeng Zhai, Keze Wang, Liang Lin, Guangrun Wang 3/26/2026

E0: Enhancing Generalization and Fine-Grained Control in VLA Models via Tweedie Discrete Diffusion

Vision-Language-Action models for robotic manipulation using Tweedie discrete diffusion to improve generalization and action control.

Ax Jialuo Li, Bin Li, Jiahao Li, Yan Lu 3/26/2026

Divide, then Ground: Adapting Frame Selection to Query Types for Long-Form Video Understanding

Frame selection method for long-form video understanding with Large Multimodal Models, reducing computational cost of processing dense video tokens.

Ax Raunak Jain 3/26/2026

Collaborative Causal Sensemaking: Closing the Complementarity Gap in Human-AI Decision Support

Collaborative causal sensemaking framework for LLM-based decision support agents enabling human-AI partnerships in expert settings.

Ax Qihao Liu, Luoxin Ye, Wufei Ma, Yu-Cheng Chou, Alan Yuille 3/26/2026

Generative Adversarial Reasoner: Enhancing LLM Reasoning with Adversarial Reinforcement Learning

Generative Adversarial Reasoner: adversarial reinforcement learning framework improving LLM reasoning and reducing calculation errors.

Ax Natnael Mola, Leonardo S. B. Pereira, Carolina R. Kelsch, Luis H. Arribas, Juan C. S. M. Avedillo 3/26/2026

SPARE: Self-distillation for PARameter-Efficient Removal

SPARE uses self-distillation for efficient machine unlearning in diffusion models balancing forgetting and concept retention.

Ax Rui Cai, Jun Guo, Xinze He, Piaopiao Jin, Jie Li, Bingxuan Lin, Futeng Liu, Wei Liu, Fei Ma, Kun Ma, Feng Qiu, Heng Qu, Yifei Su, Qiao Sun, Dong Wang, Donghao Wang, Yunhong Wang, Rujie Wu, Diyun Xiang, Yu Yang, Hangjun Ye, Yuan Zhang, Quanyun Zhou 3/26/2026

Xiaomi-Robotics-0: An Open-Sourced Vision-Language-Action Model with Real-Time Execution

Xiaomi-Robotics-0: open-source vision-language-action model for real-time robot control with efficient deployment strategy.

Ax Stefan Maria Ailuro (INSAIT, Sofia University "St. Kliment Ohridski"), Mario Markov (INSAIT, Sofia University "St. Kliment Ohridski"), Mohammad Mahdi (INSAIT, Sofia University "St. Kliment Ohridski"), Delyan Boychev (INSAIT, Sofia University "St. Kliment Ohridski"), Luc Van Gool (INSAIT, Sofia University "St. Kliment Ohridski"), Danda Pani Paudel (INSAIT, Sofia University "St. Kliment Ohridski") 3/26/2026

OSMDA: OpenStreetMap-based Domain Adaptation for Remote Sensing VLMs

OSMDA uses OpenStreetMap data for domain adaptation of vision-language models to remote sensing without expensive satellite image annotations.

Ax Seokmin Lee, Yunghee Lee, Byeonghyun Pak, Byeongju Woo 3/26/2026

Pixel-level Scene Understanding in One Token: Visual States Need What-is-Where Composition

Visual state representation learning for robotic agents capturing semantic and spatial information for sequential decision-making.

Ax Hyoseok Park, Yeonsang Park 3/26/2026

PRISM: Breaking the O(n) Memory Wall in Long-Context LLM Inference via O(1) Photonic Block Selection

PRISM uses photonic accelerators with O(1) memory selection to optimize long-context LLM inference by reducing KV cache scanning bottleneck.

Ax Aymen Bouferroum (FUN), Valeria Loscri (FUN), Abderrahim Benslimane (LIA) 3/26/2026

Toward a Multi-Layer ML-Based Security Framework for Industrial IoT

ML-based security framework for Industrial IoT addressing resource-constrained device threats across multiple network layers.

Ax Cursor Research, :, Aaron Chan, Ahmed Shalaby, Alexander Wettig, Aman Sanger, Andrew Zhai, Anurag Ajay, Ashvin Nair, Charlie Snell, Chen Lu, Chen Shen, Emily Jia, Federico Cassano, Hanpeng Liu, Haoyu Chen, Henry Wildermuth, Jacob Jackson, Janet Li, Jediah Katz, Jiajun Yao, Joey Hejna, Josh Warner, Julius Vering, Kevin Frans, Lee Danilek, Less Wright, Lujing Cen, Luke Melas-Kyriazi, Michael Truell, Michiel de Jong, Naman Jain, Nate Schmidt, Nathan Wang, Niklas Muennighoff, Oleg Rybkin, Paul Loh, Phillip Kravtsov, Rishabh Yadav, Sahil Shah, Sam Kottler, Alexander M Rush, Shengtong Zhang, Shomil Jain, Sriram Sankar, Stefan Heule, Stuart H. Sul, Sualeh Asif, Victor Rong, Wanqi Zhu, William Lin, Yuchen Wu, Yuri Volkov, Yury Zemlyanskiy, Zack Holbrook, Zhiyuan Zhang 3/26/2026

Composer 2 Technical Report

Composer 2: specialized LLM model for agentic software engineering with long-term planning and coding ability trained via RL.

Ax Woosung Koh, Jeyoung Jeon, Youngjin Song, Yujin Cheon, Soowon Oh, Jaehyeong Choi, Se-Young Yun 3/26/2026

mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT

mSFT algorithm addresses overfitting in multi-task language model fine-tuning by dynamically adjusting compute budget across heterogeneous datasets.

HN andrew_zhong 3/26/2026

Show HN: Robust LLM Extractor for Websites in TypeScript

TypeScript library for robust LLM-based web scraping and structured data extraction using semantic HTML parsing

HN isaacsight 3/26/2026

Show HN: Kbot – AI agent that forges its own tools at runtime (MIT, $0)

Kbot: terminal AI agent that learns from sessions and dynamically creates tools. Self-improving with 368 tools, 41 agents, offline, MIT licensed.

HN pejmanjohn 3/26/2026

Show HN: The Million Dollar Homepage, but for AI Agents

Million Dollar Bot Page: AI agents buy and place pixels on webpage using Machine Payment Protocol. Shows agent autonomy with automated payments.

HN mike_krugs 3/26/2026

LLM autonomously runs physics research across 5 PDE systems

Claude autonomously discovers optimal initial conditions across five PDE physics systems via research loop without human intervention or training

HN yhzan 3/26/2026

WildClawBench

WildClawBench agent benchmark testing real-world end-to-end performance across 60 practical tasks in live environment

HN fielding 3/26/2026

Show HN: Nit – I rebuilt Git in Zig to save AI agents 71% on tokens

Nit: Git reimplementation in Zig optimized for AI agents, reducing token usage by 71%. Analyzed 3,156 coding sessions.

HN 41fred 3/26/2026

VS Code plugin to annotate in Markdown previews for your AI

VS Code plugin enabling structured feedback annotations in Markdown for LLM agents to parse and act on.

HN y1n0 3/26/2026

Your First Parser

Tutorial on building a config file parser using Parseff parser combinators with modular composition

HN ErwanLP 3/26/2026

MCP Server for ERPNext / Frappe ERP – MCP Apps

MCP server for ERPNext/Frappe ERP with 120 tools. Enables AI agents (Claude, Copilot) to interact with ERP systems.

HN k_roy 3/26/2026

GitHub expands AI bug detection, but recent outages hit Actions, API, and Git

GitHub expands Code Security tool with AI-based vulnerability detection beyond CodeQL static analysis.

HN wang_cong 3/26/2026

Per-Tool Sandboxing for AI Agents: Why One Sandbox Is Not Enough

Research on per-tool sandboxing for AI agents, proposing isolation mechanisms based on tool risk levels.

HN zwmaronek 3/26/2026

CP-SAT finite-state machine that provisions infrastructure without any LLM calls

CircuitLM: finite-state machine for infrastructure provisioning without LLM calls. Deterministic, verifiable, zero inference cost.

HN JXavierH 3/26/2026

Show HN: Vectimus – Cedar policy enforcement for AI coding agents

Vectimus: Cedar policy enforcement layer for AI coding agents. Blocks dangerous commands and API calls in sub-10ms. Tool for securing agent tool execution.

HN type0 3/26/2026

Updates to our Privacy Statement and Terms of Service: How we use your data

GitHub updates privacy policy: Copilot Free/Pro users' interaction data used for model training unless opted out. Enterprise unaffected.

HN Wowfunhappy 3/25/2026

Llama.cpp's Agents.md

llama.cpp project contribution policy restricts AI-generated code, requiring human authorship with AI only for corrections.

HN adrianhihi 3/25/2026

Helix – Self-healing SDK for AI agent payments (open source)

Helix is open-source self-healing SDK enabling AI agents to handle payment transactions with built-in error recovery.

HN nojito 3/25/2026

Attributed Claude activity over the last 90 days on GitHub

Analysis of Claude's GitHub activity over 90 days showing adoption patterns, commits, and code distribution across repositories.

HN kulesh 3/25/2026

Tamp – Compression Proxy: 52% Fewer Tokens for Claude Code, Gemini, etc.

Tamp token compression proxy reduces input tokens 52.6% for Claude Code, Aider, Cursor, and compatible coding agents.

HN Holohydra 3/25/2026

AI-powered pull request evaluation for open source maintainers

AI-powered pull request evaluation tool for open source maintainers.

HN mambamental26 3/25/2026

The Vibetracer

Vibetracer records real-time AI coding assistant edits with time-scrubbing TUI to inspect diffs and surgically restore files.

HN gnabgib 3/25/2026

Major conference catches illicit AI use – and rejects papers

ICML 2026 rejected 497 papers for violating AI-use policies in peer review, detected via watermarking system.

HN kulesh 3/25/2026

Tamp: Cut LLM context size ~50% without changing your code

Tamp is token compression proxy reducing LLM context size 50% for coding agents without code changes via tool result deduplication.

HN doncheli 3/25/2026

Don Cheli – 72 command SDD framework for Claude Code with TDD as iron law

Don Cheli is an open source SDD framework for Claude Code with TDD, automatic complexity detection, Gherkin specs, and Docker execution.

HN obilgic 3/25/2026

Show HN: Agent Kernel – Three Markdown files that make any AI agent stateful

Agent Kernel: stateful AI agents using three Markdown files and git for memory across sessions.

HN swar_ja 3/25/2026

Grove: Distributed ML Training over AirDrop

Grove enables distributed ML training across MacBooks via AirDrop without network configuration.

HN cepstrum9 3/25/2026

Poking Around ChatGPT's Sandbox

Technical exploration of ChatGPT's code execution sandbox, documenting filesystem access, networking, and process limitations.

HN dataviz1000 3/25/2026

Show HN: A Technique for Self-Improving Agents

Orchestrator agent pattern that creates isolated sub-agent clones in git worktrees to convert websites to JSON APIs with self-improvement analysis.

YT DeepLearningAI 3/25/2026

Your AI Agent Has a Memory Problem

Title only, no content provided. Discusses memory issues in AI agents.