Ax Xianzhou Zeng, Jing Huang, Chunmei Xie, Gongrui Nan, Siye Chen, Mengyu Lu, Weiqi Xiong, Qixuan Zhou, Junhao Zhang, Qiang Zhu, Yadong Li, Xingzhong Xu 5/27/2026

UCPO: Uncertainty-Aware Policy Optimization

UCPO: Uncertainty-aware RL policy optimization addressing advantage bias and overconfidence in LLM training.

Ax Hayeong Lee, JunHyeok Oh, Byung-Jun Lee 5/27/2026

TABX: A High-Throughput Sandbox Battle Simulator for Multi-Agent Reinforcement Learning

TABX: High-throughput JAX-based sandbox for designing and evaluating custom multi-agent reinforcement learning scenarios.

Ax Yuxin Jiang, Yuchao Gu, Ivor W. Tsang, Mike Zheng Shou 5/27/2026

Olaf-World: Orienting Latent Actions for Video World Modeling

Olaf-World: Latent action learning for video world models with shared action semantics across contexts without action labels.

Ax Yasaman Torabi, Parsa Razmara, Hamed Ajorlou, Bardia Baraeinejad 5/27/2026

NeuroMambaLLM: Dynamic Graph Learning of fMRI Functional Connectivity in Autistic Brains Using Mamba and Language Model Reasoning

NeuroMambaLLM combines Mamba state-space models with LLM reasoning for dynamic fMRI functional connectivity analysis in autism.

Ax Valentin Dorseuil (DI-ENS), Jamal Atif (CMAP), Olivier Capp\'e (DI-ENS) 5/27/2026

Assessing Per-Sample Membership Inference Vulnerability without Retraining

Method for assessing per-sample membership inference attack vulnerability without shadow model retraining.

Ax Kang He, Kaushik Roy 5/27/2026

SWE-Adept: An LLM-Based Agentic Framework for Deep Codebase Analysis and Structured Issue Resolution

SWE-Adept: LLM-based agentic framework with two-agent architecture for deep codebase navigation and test-driven issue resolution.

Ax Kevin H. Guo, Chao Yan, Avinash Baidya, Katherine Brown, Xiang Gao, Juming Xiong, Zhijun Yin, Bradley A. Malin 5/27/2026

Stop Listening to Me! How Multi-turn Conversations Can Degrade LLM Reliability

Stick-or-switch framework evaluating how multi-turn conversations degrade LLM reliability across question-answer partitions in healthcare contexts.

Ax Shaily Roy, Harshit Sharma, Daniel A. Adler, Srijan Sen, Tanzeem Choudhury, Asif Salekin 5/27/2026

Ethical Fairness without Demographics in Human-Centered AI

Fairness methods for health ML models without demographic attributes, addressing equity across diverse people and devices.

Ax Jeonghye Kim, Xufang Luo, Minbeom Kim, Sangmook Lee, Dongsheng Li, Yuqing Yang 5/27/2026

Understanding Reasoning in LLMs through Strategic Information Allocation under Uncertainty

Information-theoretic framework explaining LLM self-correction mechanisms through separation of procedural advancement and epistemic verbalization.

Ax Xiaojie Gu, Sherry T. Tong, Aosong Feng, Sophia Simeng Han, Jinghui Lu, Yingjian Chen, Yusuke Iwasawa, Yutaka Matsuo, Chanjun Park, Rex Ying, Irene Li 5/27/2026

Omanic: Towards Step-wise Evaluation of Multi-hop Reasoning in Large Language Models

Omanic: 4-hop QA benchmark with 10,296 examples for diagnosing where LLM multi-hop reasoning fails at intermediate steps, not just final answers.

Ax Jos\'e Guilherme Marques dos Santos, Ricardo Yang, Rui Humberto Pereira, Alexandre Sousa, Br\'igida M\'onica Faria, Henrique Lopes Cardoso, Jos\'e Duarte, Jos\'e Lu\'is Reis, Lu\'is Paulo Reis, Pedro Pimenta, Jos\'e Paulo Marques dos Santos 5/27/2026

From PDF to RAG-Ready: Evaluating Document Conversion Frameworks for Domain-Specific Question Answering

Systematic evaluation of PDF-to-Markdown conversion frameworks (Docling, MinerU, Marker, DeepSeek OCR) for RAG systems; measures downstream QA accuracy across 21 pipeline configurations.

Ax Nathan Lambert, Florian Brand 5/27/2026

The ATOM Report: Measuring the Open Language Model Ecosystem

ATOM Report: comprehensive ecosystem analysis of ~1.5K open language models documenting adoption, builders, and trend of Chinese models overtaking U.S. counterparts in 2025.

Ax Xiaoke Guo, Songze Li, Zhiqiang Liu, Zhaoyan Gong, Yuanxiang Liu, Huajun Chen, Wen Zhang 5/27/2026

ASTRA: Adaptive Semantic Tree Reasoning Architecture for Complex Table Question Answering

ASTRA: adaptive semantic tree reasoning for complex table QA with LLMs; addresses table serialization bottlenecks through hierarchical schema-flexible representation.

Ax Yi Zhong, Buqiang Xu, Yijun Wang, Zifei Shan, Shuofei Qiao, Guozhou Zheng, Ningyu Zhang 5/27/2026

Chat2Workflow: A Benchmark for Generating Executable Visual Workflows with Natural Language

Chat2Workflow: benchmark for generating executable visual workflows from natural language; enables non-technical users to build reliable workflows reducing manual engineering overhead.

Ax Jin Seong, Wencke Liermann, Minho Kim, Jong-hun Shin, Soojong Lim 5/27/2026

When VLMs 'Fix' Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR

First systematic evaluation of multi-line handwritten math OCR in VLMs; develops semantic-aware metrics beyond lexical measures for educational AI systems.

Ax Zhentao Xu, Shangjin Zhang, Emir Poyraz, Yvonne Li, Ye Jin, Xie Lu, Xiaoyang Gu, Karthik Ramgopal, Praveen Kumar Bodigutla, Xiaofeng Wang 5/27/2026

Hierarchical Long-Term Semantic Memory for LinkedIn's Hiring Agent

Industrial-grade long-term semantic memory system for LinkedIn's LLM hiring agent; extracts signals from behavioral data, stores structured form, supports low-latency retrieval.

Ax Xulin Hu, Che Wang, Wei Yang Bryan Lim, Jianbo Gao, Zhong Chen 5/27/2026

Tracing the Dynamics of Refusal: Exploiting Latent Refusal Trajectories for Robust Jailbreak Detection

Identifies refusal trajectories in LLM activations across layers using causal tracing; enables robust jailbreak detection that persists against GCG attacks suppressing terminal refusal signals.

Ax Marcel Binz, Elif Akata, Abdullah Almaatouq, Mohammed Alsobay, Oleksii Ariasov, Franziska Br\"andle, David Broska, Jason W. Burton, Nuno Busch, Frederick Callaway, Vanessa Cheung, Brian Christian, Julian Coda-Forno, Can Demircan, Vittoria Dentella, Maria K. Eckstein, No\'emi \'Eltet\H{o}, Michael Franke, Thomas L. Griffiths, Fritz G\"unther, Susanne Haridi, Sebastian Hellmann, Stefan Herytash, Linus Hof, Eleanor Holton, Isabelle Hoxha, Zak Hussain, Akshay Jagadish, Elif Kara, Valentin Kriegmair, Evelina Leivada, Li Ji-An, Tobias Ludwig, Maximilian Maier, Marcelo G. Mattar, Marvin Mathony, Alireza Modirshanechi, Robin Na, Mariia Nadverniuk, Antonios Nasioulas, Surabhi S. Nath, Helen Niemeyer, Kate Nussenbaum, Sebastian Olschewski, Thorsten Pachur, Stefano Palminteri, Aliona Petrenco, Camille V. Phaneuf-Hadd, Angelo Pirrone, Manuel Rausch, Laura Raveling, Shashank Reddy, Milena Rmus, Evan M. Russek, Tankred Saanum, Kai Sandbrink, Louis Schiekiera, Johannes A. Schubert, Luca M. Schulze Buschoff, Nishad Singhi, Leah H. Somerville, Mikhail S. Spektor, Xin Sui, Christopher Summerfield, Mirko Thalmann, Anna I. Thoma, Taisiia Tikhomirova, Vuong Truong, Polina Tsvilodub, Konstantinos Voudouris, Kristin Witte, Shuchen Wu, Dirk U. Wulff, Hua-Dong Xiong, Songlin Xu, Lance Ying, Xinyu Zhang, Jian-Qiao Zhu, Eric Schulz 5/27/2026

Post-training makes large language models less human-like

Introduces Psych-201 dataset measuring behavioral alignment of LLMs to humans; finds post-training reduces alignment with human behavior across model families.

Ax Zekun Wu (University College London), Ze Wang (University College London), Seonglae Cho (Holistic AI), Yufei Yang (Imperial College London), Adriano Koshiyama (University College London), Sahan Bulathwela (University College London), Maria Perez-Ortiz (University College London) 5/27/2026

Tool Calling is Linearly Readable and Steerable in Language Models

Shows tool-calling errors in LLM agents are linearly readable in activation space, enabling detection of wrong tool selection before execution; demonstrates steering capability.

Ax Kaixiang Zhao, Bolin Shen, Yuyang Dai, Shayok Chakraborty, Yushun Dong 5/27/2026

GraphIP-Bench: How Hard Is It to Steal a Graph Neural Network, and Can We Stop It?

GraphIP-Bench evaluates model-extraction attacks against graph neural networks deployed as services and tests ownership defenses to prevent theft.

Ax Carol Xuan Long, David Simchi-Levi, Feng Zhu, Huangyuan Su, Andre P. Calmon, Flavio P. Calmon 5/27/2026

Reliability and Effectiveness of Autonomous AI Agents in Supply Chain Management

Study of autonomous AI agents in supply chains using Beer Game, identifying inference-time levers (model selection, policies, data sharing, prompt engineering) that shape performance; reasoning models exceed human performance and reduce costs 67%.

Ax Seungmin Jin 5/27/2026

RSD: A Local Triangulation Audit Primitive for Learned Vector Blocks

Local XAI audit method for learned vector embeddings using simplex memberships and weak signal comparison.

Ax Tim Tsz-Kit Lau, Weijie Su 5/27/2026

Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers

Optimizer design principle for respecting symmetry/equivariance in neural networks, applied to embeddings and MoE layers.

Ax Parsa Mazaheri, Kasra Mazaheri 5/27/2026

AgentAtlas: Beyond Outcome Leaderboards for LLM Agents

Evaluation framework for LLM agents beyond success metrics. Introduces diagnostic taxonomy for control-decision quality and trajectory analysis.

Ax Hai-Ling Lu, Yu-Yang Li, Yin-Bi Li, Cun-Shi Wang, A-Li Luo, Jun-Chao Liang, Shuo Li 5/27/2026

Spectra as Language: Large Language Models for Scalable Stellar Parameter and Abundance Inference

LLM approach for stellar parameter inference from spectroscopic data, treating spectra as language for large-scale surveys.

Ax Junyi Wu, Weijian Luo, Haoyang Zheng, Ruizhe Zhang, Guang Lin 5/27/2026

Diff-Instruct with Diffused Reward: Towards Principled One-step Generator RL

RL method for optimizing one-step text-to-image generators using diffusion models with improved reward matching.

Ax Khoi Le, Tri Cao, Phong Nguyen, Cong-Duy Nguyen, Anh Tuan Luu, Miao Chunyan, See-Kiong Ng, Thong Nguyen 5/27/2026

When In-Distribution Gains Fail: Evaluating Weak-to-Strong Reward Models under Preference Shift

Studies weak-to-strong generalization framework for reward models under distribution shift, identifying representation issues in preference learning.

Ax Steve Hanneke, Qinglin Meng, Shay Moran, Amirreza Shaeiri 5/27/2026

PAC Learning with Bandit Feedback: Sharp Sample Complexity in the Realizable Setting

Theoretical PAC learning analysis with bandit feedback in realizable setting. Pure learning theory without AI/LLM applications.

Ax Jianwei Tai 5/27/2026

Capability and Robustness Cannot Both Be Free: An Information-Theoretic Bound for Vision-Language-Action Models

Theoretical analysis of robustness-capability tradeoff in vision-language-action models for robotics, showing adversarial vulnerability in OpenVLA-7B.

BL 5/27/2026

Warp’s big bet on building open source with GPT-5.5

Warp terminal uses GPT-5.5 to orchestrate AI agents across workflows, reducing tokens by 30% and automating 90% of internal pull requests.

HN dotnot 5/26/2026

Why LLMs will be always Terrible at Software Architecture

Opinion piece arguing LLMs will remain poor at software architecture decisions due to inability to handle trade-offs and long-term consequences.

HN ammar_x 5/26/2026

DeepSWE: A contamination-free benchmark for long-horizon coding agents

DeepSWE contamination-free benchmark for long-horizon coding agents with 4 major improvements over public benchmarks including longer tasks and accurate verification.

HN xfax 5/26/2026

DeepSWE Benchmark

DeepSWE: benchmark for long-horizon coding agents on original software engineering tasks, addressing saturation in existing benchmarks like SWE-bench.

HN iamonthemission 5/26/2026

LMIM OS – an offline AI ecosystem. Voice, RAG, WhatsApp. ++ One file. 0 setup

LMIM OS: single-file offline AI ecosystem with voice, RAG, and WhatsApp integration, zero setup.

HN tusharcw 5/26/2026

Show HN: CredWork – a simple project tracking and showcasing tool

CredWork: project tracking and portfolio tool that integrates with AI agents, includes activity tracking and skill badges.

HN stan_kirdey 5/26/2026

Show HN: Clark-agent, a Rust library for LLM tool loops

Clark-agent: Rust library for building LLM tool loops with typed transcripts, tool calls, results, and stream events. AI agents framework.

HN 0x12A 5/26/2026

Show HN: Tool that hashes files into poems

Toy project that generates BLAKE3-based file fingerprints as five-line poems using LLMs with entropy injection. Not production-safe.

HN ColinEberhardt 5/26/2026

Twelve Ways to Be Wrong About AI-Assisted Coding

Analysis of misconceptions in measuring AI-assisted coding productivity, comparing metrics like lines of code versus tickets closed.

HN genesishash 5/26/2026

Show HN: Monkdev is a toolkit and methodology for coding with LLMs

Monkdev toolkit and methodology for improving code quality when working with LLMs like Claude Opus by providing better context. LLM coding framework.

HN jaydemks 5/26/2026

SafeDeps – a local dependency safety gate for Python/NPM/NuGet

SafeDeps is a local dependency vulnerability scanner for Python, NPM, and NuGet packages.

HN atulchoudhary 5/26/2026

Who Wants to Be Hired? (May 2026) – AI Engineer (Python, RAG, Agentic Workflows)

Job posting from AI engineer experienced in autonomous agents, RAG pipelines, and multi-agent orchestration.

HN Cassandra99 5/26/2026

Show HN: An LLM translator whose source is a single prompt

LLM translator app built from a single prompt, deployable in Claude Code or compatible LLMs with online demo.

HN renarl 5/26/2026

SkillOpt – Executive Strategy for Self-Evolving Agent Skills

SkillOpt: Framework for training self-evolving agent skills through rollout, reflection, editing, validation, and export.

HN NikhilKashyap 5/26/2026

Show HN: InterviewSignal – open-source AI-native technical interviews

Open-source AI-native technical interview platform for asynchronous developer screening using AI assistance.

HN braincoder 5/26/2026

FlowLink: MCP proxy blocking destructive AI agent commands

MCP proxy tool that intercepts and blocks destructive commands from AI agents before execution on production systems.

HN Gamemaster1379 5/26/2026

Show HN: I used AI to built static recompilers for 5 retro game systems

Developer built static recompilers for 5 retro game systems using AI assistance over 4 months.

HN htrp 5/26/2026

Microsoft – Agent Host Protocol for Running Agents in VS Code

Agent Host Protocol specification for running agents in VS Code with channel-based routing and Redux-like state management.

HN sosuke 5/26/2026

You Can Start Building LLM Skills Before You Know the Whole Shape

Opinion piece on building LLM automation skills incrementally without fully understanding the problem domain.

HN amitabhi 5/26/2026

Halgorithem Playground. A Playground for Halgorithem

Halgorithem Playground: Local tool for validating AI responses against trusted source material with ChatGPT integration.

HN praveenperera 5/26/2026

Show HN: Speakrs Full PyAnnotate pipeline in Rust/ONNX 20-37x times faster macOS

Speakrs: Rust/ONNX implementation of pyannote speaker diarization pipeline, 20-37x faster than Python on macOS.