6/26/2026
NASimJax: A GPU-Accelerated Policy Learning Framework for Penetration Testing
NASimJax: GPU-accelerated RL framework for training penetration testing agents with realistic network simulation.
6/26/2026
Learning from Equivalence Queries, Revisited
Learning framework for evolving models through user interaction queries; theoretical foundations for deployed systems.
6/26/2026
TransXion: A High-Fidelity Graph Benchmark for Realistic Anti-Money Laundering
TransXion benchmark for anti-money laundering detection using realistic transaction graph datasets.
6/26/2026
Closed-Loop CO2 Storage Control With History-Based Reinforcement Learning and Latent Model-Based Adaptation
Deep reinforcement learning for CO2 storage control with latent model adaptation under partial observability.
6/26/2026
Sutra: Tensor-Op RNNs as a Compilation Target for Vector Symbolic Architectures
Sutra: compiler for vector symbolic architectures that targets PyTorch neural networks with tensor operation fusion.
6/26/2026
Chebyshev Policies and the Mountain Car Problem: Reinforcement Learning for Low-Dimensional Control Tasks
Analytical solution to Mountain Car problem revealing optimal control simplicity and introducing Chebyshev policies as universal RL policy class.
6/26/2026
From One-Pass SGD to Data Reuse: Mini-Batch Scaling Laws in Sketched Linear Regression
Theoretical analysis of mini-batch scaling laws in sketched linear regression across single-pass and multi-pass SGD settings.
6/26/2026
LLMTabBench: Evaluating LLMs on Binary Tabular Classification From Zero to Few Shots
Benchmark evaluating LLM zero- and few-shot performance on binary tabular classification without labeled context examples.
6/26/2026
Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning
Deep reinforcement learning overlay for pair trading strategy in cryptocurrency markets with high volatility adaptation.
6/26/2026
Running the Gauntlet: Re-evaluating the Capabilities of Agents Beyond Familiar Environments
Benchmark for evaluating AI agent capabilities across diverse environments beyond common applications, addressing limitations of saturated performance on existing benchmarks.
6/26/2026
When to Write and When to Suppress: Route-Specialized Dual Adapters for Memory-Assisted Knowledge Editing
Parameter-efficient adapter approach for knowledge editing in LLMs using memory retrieval and dual routing mechanisms to update facts while preserving model behavior.
6/26/2026
Signature filtering: a lightweight enhancement for statistical watermark detection in large language models
Signature filtering module enhances statistical watermark detection in LLM outputs without modifying generation or embedding.
6/26/2026
The Generalization Spectrum: A Chromatographic Approach to Evaluating Learning Algorithms
Generalization Spectrum framework evaluates learning algorithms on per-sample transfer ability rather than aggregate test scores.
6/26/2026
MiniOpt: Reasoning to Model and Solve General Optimization Problems with Limited Resources
MiniOpt framework enabling LLMs to reason, model and solve diverse optimization problems with minimal training resources.
6/26/2026
Adversarial Robustness of AI-Generated Image Detectors in the Real World
Study of adversarial robustness of AI-generated image detectors, testing methods against evasion and poisoning attacks.
6/26/2026
Reconstruction Alignment Improves Unified Multimodal Models
Reconstruction Alignment (RECA) method improves unified multimodal models by leveraging visual understanding for better generation.
6/26/2026
Patent Representation Learning via Self-supervision
Self-supervised contrastive learning method for patent document representation, optimizing dropout and temperature settings.
6/26/2026
Solving Semi-Supervised Few-Shot Learning from an Auto-Annotation Perspective
Semi-supervised few-shot learning approach using vision-language models and auto-annotation for learning from limited labeled data.
6/26/2026
Metaphors are a Source of Cross-Domain Misalignment of Large Reasoning Models
Study showing metaphors in training data cause cross-domain misalignment and reasoning errors in large language models.
6/26/2026
Hierarchical Fault Detection and Diagnosis for Transformer Architectures
DEFault++ hierarchical fault detection system for identifying component-level failures in transformer architectures without visible errors.
6/26/2026
Weak-to-Strong Elicitation via Mismatched Wrong Drafts
Weak model drafts improve strong model GRPO training by injecting mathematically wrong domain-specific examples, outperforming standard on-policy RL.
6/26/2026
Evaluating Deep Research Agents on Expert Consulting Work: A Benchmark with Verifiers, Rubrics, and Cognitive Traps
Benchmark evaluating deep research agents on expert consulting tasks with 70 SME-authored prompts embedding cognitive traps.
6/26/2026
Symbolic Reasoning Frameworks Trigger Memory-Mediated Ecosystem Dynamics in Multi-Agent LLM Systems
Multi-agent LLM systems with symbolic reasoning frameworks show emergent risk behavior changes through memory interactions and reflective prompts.
6/26/2026
Representation Costs in Data Science: Foundations and the Quasi-Banach Spaces of Deep Neural Networks
arXiv paper developing framework for analyzing representation costs in parametric data-fitting and deep neural networks via regularizers.
6/26/2026
Prompt, Plan, Extract: Zero-Shot Agentic LLMs Workflows for Lung Pathology Extraction from Clinical Narratives
arXiv paper on agentic LLM workflows for zero-shot information extraction from lung pathology reports using prompt-plan-extract approach.
6/26/2026
A-Evolve-Training: Autonomous Post-Training of a 30B Model
arXiv paper on A-Evolve-Training, autonomous post-training system for 30B model with no human-in-loop over multiple weeks.
6/26/2026
GRAG: Generic Response-Augmented Generation Framework for Personalized Conversational Systems
arXiv paper on GRAG framework for personalized conversational agents with separate treatment of content grounding and personalization.
6/26/2026
How Reliable Is Your Jailbreak Judge? Calibration and Adversarial Robustness of Automated ASR Scoring
arXiv paper evaluating reliability of automated jailbreak judges for LLM safety, comparing classifiers vs human evaluation.
6/26/2026
Autodata: An agentic data scientist to create high quality synthetic data
arXiv paper introducing Autodata, an AI agent that acts as data scientist to generate synthetic training data via Agentic Self-Instruct.
6/26/2026
Building domain-specific AI chatbots with Claude
Technical guide on building domain-specific expert chatbots using Claude with architectural patterns and two-model approach.
6/26/2026
Show HN: A GPU/VRAM filter for finding LLMs that will run on your hardware
GPU/VRAM filter tool for searching which LLMs will run on specific hardware at various quantization levels.
6/26/2026
Ask HN: How has the ML engineer job evolved outside of LLM labs?
Discussion of how ML engineering roles have evolved as frontier LLMs replace custom-trained models in non-lab environments.
6/26/2026
Rura: Interactive TUI scratchpad for building shell pipelines
Terminal UI tool for building and testing shell pipelines interactively with live output and diff capabilities.
6/26/2026
White House asks OpenAI to limit its next model release
White House requests OpenAI restrict GPT 5.6 release to government partners due to advanced capabilities.
6/26/2026
Guardrails for offensive AI agents and why instructions are never enough
Discusses guardrails and safety mechanisms for preventing AI agents from executing harmful instructions.
6/26/2026
Local-first AI coding assistant for IntelliJ-based IDEs (paid)
Local-first AI coding assistant for JetBrains IDEs with offline inference capabilities.
6/26/2026
Show HN: Ludion – routing AI inference by observed WebGPU behavior
Ludion is a routing system for AI inference that optimizes WebGPU behavior and resource allocation.
6/26/2026
Anthropic Accuses Alibaba of Largest AI Distillation Attack: 28.8M Fraudulent
Anthropic accuses Alibaba of orchestrating largest AI model distillation attack with 28.8M fraudulent exchanges.
6/26/2026
DropItDown – Drop a file, get Markdown your AI agent can read (macOS)
DropItDown: macOS tool converting files to clean Markdown locally for AI agents without account or token burn.
6/26/2026
Thermodynamic Computing from Zero to One
Extropic announces breakthrough AI algorithms and hardware addressing energy efficiency as limiting factor for AI scaling.
6/26/2026
CS2-10k: A Large-Scale Egocentric Counter-Strike 2 Dataset
CS2-10k dataset: large-scale egocentric Counter-Strike 2 video with synchronized action inputs for world model training.
6/25/2026
Show HN: HoprLabs – a Python lab for prototyping AI math ideas
HoprLab: Python CLI toolkit for simulating AI training math, estimating model size, VRAM usage, training time, and config risks before actual model training.
6/25/2026
Show HN: DeepSeek Flash inverted the economics of agent products
DeepSeek Flash inverted AI agent economics: open-source model beats Sonnet on benchmarks, eliminates pricing subsidy dynamics favoring developers.
6/25/2026
New macOS malware embeds fake errors to confuse AI analysis tools
macOS malware 'Gaslight' embeds fake errors and prompt injection strings to confuse AI-assisted malware analysis tools.
6/25/2026
Export Claude.ai chats, artifacts, and visible thinking
Chrome extension for exporting Claude.ai conversations, artifacts, and visible thinking as organized ZIP files with local browser processing.
6/25/2026
JetSpec Enables Up to 9.64x Lossless LLM Inference Speedup with Up to 1000TPS
JetSpec: LLM inference optimization achieving up to 9.64x speedup and 1000 tokens-per-second throughput using speculative decoding.
6/25/2026
General LLMs outperform specialized clinical AI tools on medical benchmarks
Study comparing general LLMs versus specialized clinical AI tools on medical benchmarks. No detailed findings or data provided.
6/25/2026
Extract structured JSON from messy text with Telnyx AI Inference
Telnyx AI Inference service extracts structured JSON from unstructured text like support tickets and emails using configurable schemas.
6/25/2026
Scaling Laws, Carefully
Technical analysis of scaling laws in deep learning, their empirical foundations, and optimal compute allocation strategies.
6/25/2026
Ask HN: Have you ever given your AI agent a phone number?
Discussion: Users sharing experiences giving AI agents phone numbers and real-world tool access for task automation.