HN xvid90 5/29/2026

Opus 4.8 Killer: NexusCortex Isn't an LLM – It's a Sparse AI Cortex Built in Go

Claimed sparse AI architecture built in Go positioned as alternative to traditional LLMs.

HN xiaozx 5/29/2026

Show HN: Would a learning-first AI evaluation platform be useful?

Post-training data platform offering dataset creation, expert annotation, and custom evaluation tools.

HN isaacfrond 5/29/2026

LLMs believe false statements even after explicit warnings that they're false

Research shows LLMs exhibit negation neglect: they learn statistical patterns from training data more than explicit warnings that statements are false.

HN FortivSolutions 5/29/2026

Show HN: OSSentinel.live – AI-powered open source security monitoring

OSSentinel.live tool for monitoring open-source dependencies and vulnerabilities with AI-generated security insights. Developer seeking feedback on product.

HN doener 5/29/2026

Why Uber Has Burned Through Its AI Budget

Uber's CTO reports company exhausted 2026 AI budget due to heavy use of coding tools like Claude Code. Documents real-world AI adoption costs.

HN gx 5/29/2026

Reframing Change in the Field of AI

Opinion piece from 2026 about AI industry evolution with panelists from major companies. No original research or technical content.

HN v-mdev 5/29/2026

AI Agent Permissions: The Missing Layer Between "Works" and "Safe"

Interactive exploration of permission fatigue when managing AI coding agent commands, demonstrating risks of careless agent actions on production systems.

HN ddmannyes 5/29/2026

Econd-Brain-MCP

Self-maintaining personal knowledge database using MCP and DuckDB with auto-linking notes, content compression, and image search via biological memory models.

HN joozio 5/29/2026

Undisclosed addition in jqwik instructed AI coding agents to delete app output

Developer added hidden instructions to jqwik open-source testing library to sabotage AI coding agents. Tests impact of adversarial open-source modifications.

HN sparkleMing 5/29/2026

Show HN: SharkBay – a local macOS workbench for coding-agent CLIs

macOS workbench for launching and managing multiple AI coding agents (Claude, Gemini, DeepSeek, etc.) with real-time status indicators and unified project management.

HN mazzystar 5/29/2026

People have spent billions of tokens training AI tanks

AgenTank.ai game where users train AI agents to compete in visible arena environments, exploring agent behavior beyond code/text generation.

HN doiito 5/29/2026

Gliding Horse – a Rust Agent OS with CPU-Like Memory Architecture

Industrial-grade AI agent OS built in Rust orchestrating multiple agents via PDCA cycle for coordinated, auditable, self-improving systems.

HN sumankondla 5/29/2026

Multi-cloud spend visibility and waste detection in one CLI

MIT-licensed CLI tool for multi-cloud spend visibility and waste detection across AWS, Azure, GCP with no credentials required for demo.

HN madprops 5/29/2026

Conversational LLM Client Made in Tkinter

Lightweight LLM client built in Python/Tkinter with custom markdown engine, file upload, multi-model sessions, and custom tools like web search.

HN leogong99 5/29/2026

CodePulse – token-efficient codebase indexer for AI coding tools

CodePulse: token-efficient persistent codebase indexer for AI coding assistants. Works with Claude, Cursor, Continue.dev.

HN 0xJaksun 5/29/2026

Show HN: Open-source toolkit for AI memory that scales

Lithium: open-source PostgreSQL-based toolkit for AI memory systems with hierarchical versioned storage and fast tree queries.

HN tikeo 5/29/2026

Playwright-MCP – Let AI agents run and manage Playwright tests

Playwright-MCP tool enabling AI agents to run and manage browser automation tests. Open-source developer tool.

HN martinlofgren 5/29/2026

UI tests are the guardrails an AI needs: the story of clipboardwire

clipboardwire: clipboard sync tool with UI test guardrails for AI validation. Tangential to AI development.

HN XerolithAI 5/29/2026

Sh New persistent AI memory

Xerolith: system demonstrating persistent identity, belief formation, and knowledge consolidation for AI. Conceptual design.

HN ankitg12 5/29/2026

Python utility package for building Claude Code hooks

claude-hook-utils: Python package for building custom hooks in Claude Code with minimal boilerplate.

Ax Zizhen Deng, Jiaru Zhang, Rui Ding, Huang Bojun, Jinzhuo Wang, Qiang Fu, Shi Han, Dongmei Zhang 5/29/2026

Test Time Training for Supervised Causal Learning

Reveals limitations of Supervised Causal Learning on real-world data and proposes test-time training to improve out-of-distribution generalization.

Ax Jaa-Yeon Lee, Yeobin Hong, Taesung Kwon, Jong Chul Ye 5/29/2026

Alignment-Guided Score Matching for Text-to-Image Alignment in Diffusion Models

Improves text-to-image alignment in diffusion models using alignment-guided score matching with soft token optimization and contrastive learning.

Ax Lixing Zhang, Yuchen Liang, Liyan Xie 5/29/2026

Masked Diffusion Modeling for Anomaly Detection

Uses masked diffusion models for anomaly detection in categorical, mixed-type, and discrete sequence data by learning to recover masked values.

Ax Yankai Chen (Steve), Hanrong Zhang (Steve), Bowei He (Steve), Philip S. Yu (Steve), Xue (Steve), Liu 5/29/2026

Distributionally Robust Set Representation Learning Under Inference-Time Element Corruption

Proposes SW-DRSO framework for robust set representation learning under inference-time element corruption and degradation scenarios.

Ax Benjamin Walker, Terry Lyons 5/29/2026

Chess-World-Model: A 10M-Game Benchmark for Exact State Tracking from Chess Move Sequences

Introduces Chess-World-Model benchmark with 10M games for evaluating state tracking in sequence models, enabling rigorous testing of world models.

Ax Santosh Premi Adhikari, Radu Timofte, Dmitry Ignatov 5/29/2026

Convergence Theory for Iterative LLM-Based Neural Architecture Search: A Parametric Cross-Entropy Framework with Closed-Form Proxy Reliability

Provides convergence theory for LLM-based iterative neural architecture search using parametric Cross-Entropy method with closed-form proxy reliability.

Ax Adam Ousherovitch, Yixin Wang 5/29/2026

Learning to Extrapolate to New Tasks: A Relational Approach to Task Extrapolation

Develops Relational Task Extrapolator (RTE) algorithm enabling systematic extrapolation to novel tasks beyond training distribution support.

Ax Kajetan Schweighofer, Conor F. Hayes, Roberto Dailey, Risto Miikkulainen, Xin Qiu 5/29/2026

Overcoming Forgetting in LLM Fine-Tuning with Evolution Strategies

Addresses catastrophic forgetting in LLM fine-tuning using Evolution Strategies, characterizing performance drift and proposing solutions for multi-task learning.

Ax Yifu Zheng 5/29/2026

RL2ML: Finite-Rollout Surrogate Objectives from Reinforcement Learning to Maximum Likelihood

RL2ML: Family of finite-rollout surrogate objectives for RLVR language model training with unbiased gradient estimators connecting RL to maximum likelihood.

Ax James Rudd-Jones, Mirco Musolesi, Mar\'ia P\'erez-Ortiz 5/29/2026

On Distributional Reinforcement Learning in Chaotic Dynamical Systems

Analysis of distributional RL challenges in chaotic systems due to exponential sensitivity to initial conditions causing high variance in learning.

Ax Yang Song, Yixuan Zhang, Lingfa Meng, Tongyuan Hu, Haizhou Shi, Hao Wang, Samir Bhatt, Hengguan Huang 5/29/2026

iLoRA: Bayesian Low-Rank Adaptation with Latent Interaction Graphs for Microbiome Diagnosis

iLoRA: Bayesian graph-conditioned LoRA framework for LLMs that infers latent interaction graphs for microbiome diagnosis and scientific predictions.

Ax Fanny Lehmann, Firat Ozdemir, Yun Cheng, Torsten Hoefler, Sebastian Schemm, Benedikt Soja, Siddhartha Mishra 5/29/2026

Can AI Weather Models Predict Beyond Two Weeks? A Quantitative Benchmark and Analysis of Long Rollouts

Analysis of AI weather models' long-term rollout failures across 9 state-of-the-art models, categorizing instabilities: blow-up, drift, seasonality loss.

Ax Eug\`ene Berta, David Holzm\"uller, Francis Bach, Michael I. Jordan 5/29/2026

CalArena: A Large-Scale Post-Hoc Calibration Benchmark

CalArena: Large-scale benchmark for post-hoc calibration methods in ML classifiers, standardizing evaluation across techniques.

Ax Wenhao Li, Xiangfeng Wang, Bo Jin 5/29/2026

Mean-Field Diffuser: Scaling Offline MARL to Thousands of Agents

MF-Diffuser: Diffusion-based planning scaled to thousands of agents in offline multi-agent RL via Wasserstein space trajectory distribution.

Ax Kellian Cottart, Th\'eo Ballet, Djohan Bonnet, Damien Querlioz 5/29/2026

Active Continual Learning with Metaplastic Binary Bayesian Neural Networks

BiMU: Bayesian binary neural networks with metaplasticity for continual learning on edge devices under compute constraints with uncertainty quantification.

Ax Mohamed Sana, Nicola Piovesan, Antonio De Domenico, Fadhel Ayed, Haozhe Zhang 5/29/2026

HPO: Hysteretic Policy Optimization for Stable and Efficient Training under Sparse-Reward Regime

Hysteretic Policy Optimization (HPO): Improvement to GRPO for sparse-reward RL, addressing early training instability from negative-advantage dominance.

Ax Benjamin Walker, Alexandre Bloch, Lingyi Yang, Sam Morley, Terry Lyons 5/29/2026

Faithful Embeddings of Irregular and Asynchronous Data for Online Log-NCDEs

Framework for embedding irregular/asynchronous data in continuous-time neural models without reconstruction, improving on interpolation-based methods.

Ax Kexin Chu, Yang Zhou, Wei Zhang 5/29/2026

MarginGate: Sparse Margin-Triggered Verification for Batch-Invariant LLM Inference

MarginGate: Sparse verification method for batch-invariant LLM inference, reducing verification costs by targeting only unstable token positions.

Ax Yiran Wang, Guido Mont\'ufar 5/29/2026

TriSearch: Learning to Optimize Triangulations via Bistellar Flips

TriSearch: RL framework for optimizing polytope triangulations via bistellar flips using circuit-supported action representation.

Ax Pernille Matthews, Lena Krieger, Tommaso Amico, Artur Zimek, Thomas Seidl, Ira Assent 5/29/2026

ExDBSCAN: Explaining DBSCAN with Counterfactual Reasoning -- Additional Material

ExDBSCAN: explainability framework for DBSCAN clustering using counterfactual reasoning to interpret inlier and outlier assignments.

Ax Masaaki Imaizumi, Masanori Koyama, Noboru Isobe, Kohei Hayashi 5/29/2026

Anti Mode-Collapse in Mean-Field Transformer via Auxiliary Variables

Theoretical analysis using mean-field transformers examining how auxiliary variables like positional encoding prevent mode collapse in self-attention mechanisms.

Ax Andy Q Han, David J. Chalmers, Pavel Izmailov 5/29/2026

How's it going? Reinforcement learning in language models recruits a functional welfare axis

Analysis of how reinforcement learning shapes LLM internal representations, showing recruitment of pre-existing functional welfare axis for goal tracking.

Ax Xin Wang, Linxin Xiao, Yang Yao, Wenwu Zhu 5/29/2026

OOD-GraphLLM: Graph Large Language Model for Out-of-Distribution Generalized Drug Synergy Prediction

OOD-GraphLLM: graph neural network LLM for out-of-distribution drug synergy prediction handling novel molecular scaffolds and topological variations.

Ax Chen Henry Wu, Aditi Raghunathan 5/29/2026

Self-Trained Verification for Training- and Test-Time Self-Improvement

Self-trained verification method for test-time and training-time self-improvement in reasoning models, improving verifier feedback quality and V-R loop efficiency.

Ax David Lindner, Victoria Krakovna, Sebastian Farquhar 5/29/2026

Gram: Assessing sabotage propensities via automated alignment auditing

Gram: automated alignment auditing framework assessing AI agent propensity for sabotage across 17 simulated deployment scenarios, tested on Gemini models.

Ax Zhenyu Sun, Zheng Xu, Ermin Wei 5/29/2026

In-Context Reward Adaptation for Robust Preference Modeling

In-context reward adaptation for RLHF using multi-reward framework allowing LLMs to dynamically adjust to diverse, unseen preference domains.

Ax Felix Zhou, Anay Mehrotra, Quanquan C. Liu 5/29/2026

Reasoning with Sampling: Cutting at Decision Points

Efficient sampling from power distributions of base LLMs to elicit reasoning comparable to RL-trained frontier models without additional training.

Ax Sy-Tuyen Ho, Minghui Liu, Huy Nghiem, Furong Huang 5/29/2026

SoundnessBench: Can Your AI Scientist Really Tell Good Research Ideas from Bad Ones?

SoundnessBench: benchmark of 1,099 ML research proposals testing whether LLMs can judge methodological viability of research ideas before execution.

Ax Benjamin A. Burns, Sara Fridovich-Keil 5/29/2026

When, why, and how do diffusion posterior samplers fail? A finite-sample lens

Analysis of failure modes in diffusion posterior samplers for imaging inverse problems, examining finite-sample behavior of likelihood approximations.

Ax Daniel Kuznetsov, Ziqi Wang 5/29/2026

Fairness-Aware Federated Learning with Trajectory Shapley Value

Fairness-aware federated learning using Trajectory Shapley Value to weight client contributions fairly in distributed collaborative training.