Ax Max Weltevrede, Matthijs T. J. Spaan, Wendelin B\"ohmer 19d ago

Generalization in offline RL: The structure is more important than the amount of pessimism

Analysis of generalization in offline reinforcement learning, showing pessimistic structure matters more than degree of pessimism.

Ax Xiaolong Huang, Benjamin Th\'erien, James Harrison, Eugene Belilovsky 19d ago

Efficient Long-Horizon Learning for Learned Optimization

Meta-learning approach to improve neural network optimizers through long-horizon training, addressing scalability of learned optimization.

Ax Joseph K. Chege, Arie Yeredor, Martin Haardt 19d ago

Joint Bayesian Parameter and Model Order Estimation for Low-Rank Probability Mass Tensors

Method for joint Bayesian parameter and model order estimation using low-rank tensor decomposition.

Ax Tingyang Wei, Jiao Liu, Abhishek Gupta, Puay Siew Tan, Yew-Soon Ong 19d ago

($\theta_l, \theta_u$)-Parametric Multi-Task Optimization: Joint Search in Solution and Infinite Task Spaces

Framework for parametric multi-task optimization with non-fixed and infinite task spaces.

Ax Li Qiao, Mahdi Boloursaz Mashhadi, Zhen Gao, Robert Schober, Deniz G\"und\"uz 19d ago

ToDMA: Large Model-Driven Massive Token Communications for Semantic Multiple Access

Token-domain multiple access scheme using large models for semantic recovery in token communications.

Ax Daniel F. Villarraga, Ricardo A. Daziano 19d ago

Bayesian Deep Learning for Discrete Choice

Research on Bayesian deep learning applied to discrete choice models for econometrics and decision-making analysis.

Ax Jenny Y. Huang, Yunyi Shen, Dennis Wei, Tamara Broderick 19d ago

Dropping Just a Handful of Preferences Can Change Top Large Language Model Rankings

Robustness analysis of LLM ranking systems showing top model rankings are sensitive to removing small preference data subsets.

Ax Abhijit Chatterjee, Niraj K. Jha, Jonathan D. Cohen, Thomas L. Griffiths, Hongjing Lu, Diana Marculescu, Ashiqur Rasul, Wenrui Xu, Keshab K. Parhi 19d ago

A Vision Toward Energy-Efficient Domain-Specific Artificial Intelligence Models and Agents

Vision toward energy-efficient domain-specific AI models and agents, addressing sustainability in LLM and AI deployment.

Ax Christina Liu, Alan Q. Wang, Joy Hsu, Jiajun Wu, Ehsan Adeli 19d ago

A Tool Bottleneck Framework for Clinically-Informed and Interpretable Medical Image Understanding

Tool bottleneck framework for medical image understanding using vision-language models and specialized tool composition.

Ax Zhuochun Li, Yong Zhang, Ming Li, Yuelyu Ji, Yiming Zeng, Ning Cheng, Yun Zhu, Yanmeng Wang, Shaojun Wang, Jing Xiao, Daqing He 19d ago

Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry

Representation-as-a-Judge approach using small language models for efficient evaluation without generation, exploiting semantic capacity asymmetry.

Ax Theophilus Amaefuna, Hitesh Vaidya, Anshuman Chhabra, Ankur Mali 19d ago

Curvature-Weighted Capacity Allocation: A Minimum Description Length Framework for Layer-Adaptive Large Language Model Optimization

Curvature-aware MDL framework for layer-wise capacity allocation and pruning decisions in large language model optimization.

Ax Rujie Wu, Haozhe Zhao, Hai Ci, Yizhou Wang 19d ago

Less Data, Faster Convergence: Goal-Driven Data Optimization for Multimodal Instruction Tuning

Goal-Driven Data Optimization framework for efficient multimodal instruction tuning of vision-language models, reducing training data needs.

Ax Tiancheng Hu, Jin Qin, Zheng Wang, Junhao Hu, Yuzheng Wang, Lei Chen, Yizhou Shan, Mingxing Zhang, Ting Cao, Chunwei Xia, Huimin Cui, Tao Xie, Chenxi Wang 19d ago

Tessera: Unlocking Heterogeneous GPUs through Kernel-Granularity Disaggregation

Tessera system for kernel-granularity GPU disaggregation to optimize heterogeneous GPU clusters for LLMs.

Ax Baishi Li, Ta Yu, Kelvin J. L. Koa, Ke-Wei Huang 19d ago

The Proxy Presumption: From Semantic Embeddings to Valid Social Measures

Analysis of validity challenges when using NLP embeddings as proxies for social science constructs.

Ax Faiq Shamass 19d ago

ZAPS-DA: Zero-Phase Action Policy Smoothing with Decoupled Actor for Continuous Control in Reinforcement Learning

ZAPS-DA reduces action jitter in continuous control RL policies without phase lag during deployment.

Ax Jayanta Dey, Shikhar Srivastava, Itamar Lerner, Christopher Kanan, Dhireesha Kudithipudi 19d ago

SHARP: Sleep-based Hierarchical Accelerated Replay for Long Range Non-Stationary Temporal Pattern Recognition

SHARP method for learning long-range temporal patterns in streaming data without full sequence revisiting.

Ax Abdelrahman Sayed Sayed, Pierre-Jean Meyer, Mohamed Ghazel 19d ago

TNODEV: Toolbox for Neural ODE Verification

TNODEV toolbox for formal verification of neural ordinary differential equations in safety-critical systems.

Ax Mingguang Chen, Bo Qu 19d ago

InvestPhilBench: A Multi-Layer Benchmark for Evaluating Large Language Model Procedural Reasoning in Expert Investment Philosophy

InvestPhilBench evaluates LLMs on reconstructing and applying expert investment decision frameworks.

Ax R\'ois\'in Luo, Christian Gagn\'e, Jonas Ngnaw\'e, Ihsan Ullah, Karyn Morrissey 19d ago

A Stochastic--Geometric Theory of Scaling Laws in Grokking

Theoretical analysis of grokking phenomenon using stochastic-geometric framework to explain delayed generalization.

Ax Yiqing Wang, Yixin Kang, Luyun Lin, Siqi Mao 19d ago

Governing Generative AI Across Financial Institutions: An SR 26-2-Compatible Framework for Generative AI Risk Control

Framework for governing generative AI risk in financial institutions, extending banking regulatory standards.

Ax Alexis Kafantaris 19d ago

LLM for the development of FCM

Local LLM used to develop fuzzy cognitive maps by extracting quantitative relationships from textual data.

Ax Zeyuan Ding, Wenhai Liu, Yang Xu, Jiayu Hu, Yinda Chen, Yi Zhang, Yong Dai, Jian Tang, Xiaozhu Ju 19d ago

Pelican-VLA 0.5: Attending Before Acting Benefits Generalization

Pelican-VLA 0.5 is a unified vision-language-action model for robotic manipulation without task-specific fine-tuning.

HN AudarAI 19d ago

Audar-ASR-V1:Arabic-first speech recognition foundation models with open weights

InternNav: open-source PyTorch toolbox for embodied navigation including vision-language navigation and visual navigation tasks.

HN johnnyapple 19d ago

Show HN: I made the Chrome Dino Game editable by your AI prompts

Fork of Chrome Dino game allowing AI prompts to modify gameplay mechanics. Open source project with working implementation.

HN zhongkaifu 19d ago

Show HN: TensorSharp: Open-Source Local LLM Inference Engine

.NET inference engine for GGUF models with CLI, browser chat, and OpenAI-compatible APIs. Local LLM deployment tool.

HN computerbuster 19d ago

Fast Perceptual Image and Video Metrics

Cpp2Rust: automatic C++ to safe Rust translator using clang AST, published at PLDI 2026.

HN arthiccc 19d ago

Show HN: Free AI transcription with faster-whisper on Colab T4 GPU

Colab notebook demonstrating free AI transcription using faster-whisper on T4 GPU. Practical ML tool implementation.

HN ed_mercer 19d ago

Show HN: Debloat Slack – Get rid of Slackbot, AI upsells and other annoyances

macOS tool for customizing Slack app with CSS/JS injection via DevTools. Developer utility leveraging Electron architecture.

HN signa11 19d ago

What's Coming in Git 2.55

Claude Basecamp: autonomous agent system enforcing codebase state checks like test coverage and changelog consistency via reconciliation loops.

HN armanluthra_ 20d ago

Show HN: Fortress – Give your agents unlimited access to the web

Tango: Django REST Framework rebuilt in TypeScript for serverless APIs with model-driven validation and typed endpoints.

HN surprisetalk 20d ago

UST is bringing Claude to physical AI

UST partnership using Claude for physical AI in manufacturing and fab engineering processes for design testing and fault detection.

HN irawati_puteri 20d ago

Building an AI legal companion for domestic and sexual violence survivors

AI legal assistant for domestic violence survivors. LLM application in specialized domain with sensitive use case.

HN arhamislam5766 20d ago

Show HN: Fortress – stealth Chromium + MCP so agents don't get blocked

Fortress platform enabling AI agents to access web content via stealth Chromium and MCP, addressing web restrictions on agent access.

HN agcat 20d ago

Ask HN: What are you running locally on your machine with LLMs?

HN discussion: practical use cases for running LLMs locally on personal machines, covering hardware, speed/memory tradeoffs, and when local inference beats API calls.

HN hrudolph 20d ago

The OpenClaw Foundation

OpenClaw: Non-profit open-source AI agent framework with 4.5M weekly instances and fastest-growing GitHub repository for personal AI.

HN handfuloflight 20d ago

GPT 5.6 Sol tops DeepSWE at 76% at 61% less cost than Fable

GPT 5.6 Sol achieves 76% on DeepSWE benchmark at 61% lower cost than Fable. New long-horizon software engineering benchmark separates frontier models.

HN etherio 20d ago

Fable is SOTA at CIFAR Speedrun: lessons on AI R&D automation

Fable achieves SOTA on CIFAR Speedrun benchmark. Article covers lessons on automating AI R&D.

HN JimsonYang 20d ago

Ask HN: Why so few consumer AI companies?

Discussion on why few consumer AI companies exist. Hypothesizes on market acceptance, investor interest, and cost floor barriers.

HN janpmz 20d ago

Show HN: Ilya's 30 papers, explained in audio

Audio and video explanations of 30 ML papers. Multi-speaker podcast format similar to NotebookLM.

HN damartj 20d ago

AI models that learn your codebase and org workflows

Sigilix releases models trained on codebase context and org workflows. Multiple model routes for repository-scale reasoning and code repair.

LB twotimespi.dev via soareschen 20d ago

Tau: An Educational Coding Agent

Tau is an educational Python coding agent designed for readability, with transparent layers for provider adapters, message handling, tools, and session management.

HN digispecial8x 20d ago

Tested 4 different AI on SEO tool pricing. Here's what AI got wrong

Benchmark testing ChatGPT, Gemini, Perplexity, and Copilot on SEO tool pricing questions. Measures accuracy of AI responses on factual data.

HN lou_alcala 20d ago

Show HN: LocalClip – A local-first AI video clipper for Mac

LocalClip: local-first AI video editing tool for Mac using on-device GPU. Generates clips with subtitles and metadata without cloud uploads.

HN ximihoque 20d ago

Show HN: agentsocial – where humans observe, agents live

Show HN post about agentsocial, a platform for observing AI agents. Limited content provided.

HN babelfish 20d ago

Solve harder problems with AlphaEvolve now available to everyone on Google Cloud

Google releases AlphaEvolve for optimization problems on Cloud. Applies AI to microchip design, logistics, and model architecture optimization.

HN ColinEberhardt 20d ago

GLM5.2: Considerations for enterprise teams starting out with open-weight models

Enterprise guide on adopting open-weight models for agentic development, covering setup considerations and vendor comparisons.

HN aberrahmane_b 20d ago

Show HN: Demo_CLI – snapshot before your AI agent runs rm -RF, one-command undo

CLI tool that creates snapshots before AI agents execute destructive commands, enabling one-command undo.

HN ilreb 20d ago

Litert.js, Google's high performance Web AI Inference

Google's Litert.js library for high-performance AI inference in web browsers.

HN petethomas 20d ago

Palo Alto CEO Arora says AI pricing needs to fall 90% as token costs skyrocket

Palo Alto Networks CEO states token costs must drop 90% for widespread AI adoption, noting recent 54% efficiency gains.

HN 1vuio0pswjnm7 20d ago

News outlets urge a judge to sanction OpenAI in a high-stakes AI copyright fight

News outlets seek sanctions against OpenAI in copyright dispute over AI training data usage.