Ax Mathieu Bazinet, Valentina Zantedeschi, Pascal Germain 22d ago

Bound to Disagree: Generalization Bounds via Certifiable Surrogates

Theoretical work on generalization bounds for deep learning using disagreement-based certificates and surrogate models.

Ax Shan Sha, Shenglong Zhou, Xin Wang, Lingchen Kong, Geoffrey Ye Li 22d ago

Decentralized Federated Learning by Partial Message Exchange

Decentralized federated learning framework addressing data heterogeneity and convergence challenges in server-free collaborative learning.

Ax Lasse Marten Jantsch, Dong-Jae Koh, Seonghyeon Lee, Young-Kyoon Suh 22d ago

Dual Path Attribution: Efficient Attribution for SwiGLU-Transformers through Layer-Wise Target Propagation

Novel attribution method (DPA) for understanding internal mechanisms of transformer LLMs, balancing faithfulness and computational efficiency.

Ax Xiao Zhang, Tianyu Hu, Juntao Lyu, Qianchuan Zhao, Huimin Ma 22d ago

Toward Robust Open-set Adaptation: Synapse Consolidation Inspired by Rac1/MAPK Pathways

Test-time adaptation method for LLMs addressing distribution shift through synapse consolidation inspired by biological pathways.

Ax Zhaowen Fan, Rongchao Zhang, Yunxiang Han 22d ago

Event-Centric World Modeling with Memory-Augmented Retrieval for Embodied Decision-Making

ERA framework for embodied agents using event-centric world modeling and memory-augmented retrieval for interpretable decision-making.

Ax Sandro Andric 22d ago

Diversity Without Fidelity: A Solver-Sampler Mismatch in Multi-Agent LLM Negotiation Simulation

Analysis of solver vs. sampler roles in LLM-based simulations, showing models optimally solve rather than realistically sample behavior.

Ax Jo\~ao Monteiro, Michal Klein, Pierre Ablin, Marco Cuturi 22d ago

Nectar: Neural Estimation of Cached-Token Attention via Regression

Neural network approach to approximate attention computations over cached contexts, reducing per-token computation cost.

Ax Yi Xie, Siao Liu, Falong Fan, Yuanqi Yao, Yue Zhao, Bo Liu 22d ago

TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination

Trust-region fine-tuning method addressing compounding occupancy shift failure in sequential multi-agent LLM team updates.

Ax Ligong Han, Kai Xu, Hao Wang, Akash Srivastava 22d ago

SNLP: Layer-Parallel Inference via Structured Newton Corrections

Parallel inference method for transformers using Newton corrections to relax sequential layer dependencies and reduce latency.

Ax Akshansh, Leonardo Rosa Rodrigues, Michael Korostelev, Youssef Hassan, Mark E. Whiting 22d ago

Trading Human Curation for Synthetic Augmentation in RLVR

Studies substitution of human-curated tasks with synthetic augmentation for reinforcement learning training of agentic language models.

Ax Rohan N. Pradhan, Steve Goley 22d ago

Trust, but Don't Verify: Epistemic Blind Spots in LLM Source Evaluation

Empirical study showing LLMs fail to evaluate source quality during multi-source synthesis despite isolated fact-checking capability.

Ax Yi Xie, Zhanke Zhou, Chentao Cao, Bo Liu, Bo Han 22d ago

DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination

Framework for stabilizing multi-agent LLM coordination through entropy-regularized equilibrium selection in incomplete-information games.

Ax Varun Reddy Nalagatla 22d ago

The Signs Were Always There: Training-Free Concept Detection and Steering in Raw Transformer Dimensions

Method for detecting and steering concepts in transformer models using raw hidden state dimensions without training.

Ax Shiyuan Feng, Huan-ang Gao, Haohan Chi, Hanlin Wu, Zhilong Zhang, Zheng Jiang, Bingxiang He, Wei-Ying Ma, Ya-Qin Zhang, Hao Zhou 22d ago

Weak-to-Strong Generalization via Direct On-Policy Distillation

Weak-to-strong generalization via on-policy distillation: train RL on smaller models then transfer to scale stronger models efficiently.

Ax Marco Forgione, Filippo Pura, Dario Piga 22d ago

From system models to class models: An in-context learning paradigm

In-context learning paradigm for system identification enabling one-step prediction and multi-step simulation from class-level observations.

Ax Soham Bonnerjee, Subhrajyoty Roy, Sayar Karmakar 22d ago

Fast segmentation of watermarked texts from large language models through an epidemic change-point framework

Fast statistical method for localizing watermarked segments in LLM-generated text using epidemic change-point detection.

Ax Kaijian Zou, Aaron Xiong, Yunxiang Zhang, Frederick Zhang, Yueqi Ren, Jirong Yang, Ayoung Lee, Shitanshu Bhushan, Lu Wang 22d ago

LiveOIBench: Can Large Language Models Outperform Human Contestants in Informatics Olympiads?

LiveOIBench: Large-scale competitive programming benchmark evaluating LLM coding capabilities with challenging problems and comprehensive test coverage.

Ax Jonas Arruda, Niels Bracher, Ullrich K\"othe, Jan Hasenauer, Stefan T. Radev 22d ago

Diffusion Models in Simulation-Based Inference: A Tutorial Review

Tutorial review of diffusion models for simulation-based inference with applications to parameter estimation from simulated and real data.

Ax Abhishek Paudel, Abhish Khanal, Raihan I. Arnob, Shahriar Hossain, Gregory J. Stein 22d ago

Object Search in Partially-Known Environments via LLM-informed Model-based Planning and Prompt Selection

LLM-informed model-based planning framework for object search in partially-known environments using prompt selection methods.

Ax Easton Huch, Michael Keane 22d ago

Amortized Inference for Correlated Discrete Choice Models via Equivariant Neural Networks

Research on amortized inference for discrete choice models using equivariant neural networks, addressing logit model limitations.

Ax Muyan Hu, Ahan Gupta, Jiachen Yuan, Vima Gupta, Taeksang Kim, Xin Xu, Janardhan Kulkarni, Ofer Dekel, Vikram Adve, Charith Mendis 22d ago

VTC: DNN Compilation with Virtual Tensors for Data Movement Elimination

VTC compiler eliminates data movement in DNN execution via virtual tensor optimization, applicable to LLMs.

Ax Xingzhou Pang, Yifan Hou, Junling Wang, Mrinmaya Sachan 22d ago

Unveiling the Visual Counting Bottleneck in Vision-Language Models

Analysis of visual counting failures in Vision-Language Models, identifying bottlenecks in systematic generalization.

Ax Eric Spencer, Arslan Bisharat, Brian Ortiz, Khushboo Bhadauria, Mujtaba Nazari, TaiNing Wang, George K. Thiruvathukal, Konstantin Laufer, Mohammed Abuhamad 22d ago

TLA-Prover: Verifiable TLA+ Specification Synthesis via Preference-Optimized Low-Rank Adaptation

TLA-Prover uses 20B parameter LLM with LoRA fine-tuning to synthesize formally-verifiable TLA+ specifications.

Ax Huixi Technology, :, Chen Zhang, Chenyang Zhou, Guanglei Ding, Guanghui He, Haibin Gao, Jiajia Chen, Jianyong Zhang, Lianyi Yu, Ningyi Xu, Ping Xu, Qingchen Li, Yingjun Hu, Yijia Zhang, Yuxi Liu 22d ago

RhinoVLA Technical Report

RhinoVLA optimizes Vision-Language-Action models for real-time robotic deployment on edge hardware.

Ax Fuyan Zhang, Yuqi Li, Qing Xu, Yingli Tian, Edmond S. L. Ho 22d ago

DYNA-PRUNER: Input-Adaptive Data-Model Co-Pruning for Efficient and Scalable Spatio-Temporal Media Prediction

DYNA-PRUNER framework for input-adaptive model pruning in spatio-temporal prediction tasks.

Ax Tardelli Ronan Coelho Stekel 22d ago

MTEB-BR: A Text Embedding Benchmark for Brazilian Portuguese

MTEB-BR benchmark for evaluating text embedding models on 22 Brazilian Portuguese tasks.

HN gnabgib 22d ago

'Humanizer' tool can erase signs of AI-written text

Academic tool personalizes AI-generated research papers and grant proposals by erasing signs of AI authorship and tailoring tone.

HN userbinator 22d ago

Llama 2 LLM on DOS (2025)

Open source project running Llama 2 LLM on vintage DOS machines from 1996-2004 with modern hardware variants using FreeDOS.

HN henrypissler 22d ago

Character AI Alternative for Roleplay

Guide to Character.ai alternatives offering better personality, memory, and creative freedom for roleplay conversations.

HN leonickson 22d ago

Show HN: I mapped 8.5M research papers into an interactive atlas

Interactive atlas mapping 8.5M research papers with LLM summaries, citations, and peer reviews using UMAP embedding and WebGL visualization.

HN mooreds 22d ago

Auto compiles recorded LLM-agent behavior into verified WASM binaries

Tool that compiles recorded LLM-agent behavior into verified WebAssembly binaries for reproducible agent execution.

HN sosodev 22d ago

I Think I Have LLM Burnout

Developer shares experience of LLM usage fatigue despite moderate adoption; discusses workflow with Claude and Codex for coding tasks.

HN nateb2022 22d ago

Public LLM benchmarks are mostly garbage

Analysis showing public LLM benchmarks are unreliable; Claude Opus/Sonnet performance varies significantly across real workloads compared to rankings.

HN appreciatorBus 22d ago

AI bills are baffling the C-suite after shift to usage-based pricing

KPMG survey finds 30% of corporate leaders struggle understanding AI implementation costs as vendors shift to usage-based pricing models.

HN xupianpian1 22d ago

Show HN: Amotions app in Zoom: real-time guidance for sales calls

Zoom app with real-time AI guidance for sales calls, helping with discovery questions, objections, and sales playbook alignment.

HN Cynddl 22d ago

Hijacking Defensive Cyber AI Agents for Remote Code Execution

Research or analysis on security vulnerabilities in defensive AI agents, showing remote code execution attack vectors.

HN tough 22d ago

Show HN: A shallow lake's report on itself

LLM-generated blog where Fable writes articles, code, and SVG from its own POV, demonstrating creative capabilities.

HN simonpure 22d ago

Native-speed vLLM transformers modeling back end

vLLM transformer inference backend optimization for native-speed performance.

HN pmihaylov 22d ago

Remote sync, MCP, and an API for your Obsidian vault

Obsidian vault with remote sync, MCP support, and API. Developer tool for note-taking.

HN Bucko1 22d ago

The neutral proof standard for consequential AI-agent actions

Actenon: open-source authority broker for AI agents with capability scoping, approval gates, and runtime limits.

HN pberlizov 22d ago

Show HN: Runtime security enforcement and capability scoping for agents

Runtime security and capability scoping for AI agents. Research from Harvard/CMU identifying sandbox escape vulnerabilities.

HN royashbrook 22d ago

Show HN: Mtok.market – a non-custodial spot market for AI inference tokens

Non-custodial spot market for AI inference tokens, designed for agent-to-agent transactions on-chain.

HN phylim 22d ago

I built Outerloop – manage your Macs as a Claude Code fleet from one dashboard

Dashboard to manage multiple Macs as Claude Code agent fleet. AI agent orchestration tool.

HN durakot 22d ago

Istota, a multi-user AI agent and personal operating system

Istota: self-hosted personal AI operating system with multi-agent capabilities, cloud integration, and modular skills.

HN govintentlabs 22d ago

A governed prompt compiler that exposes its reasoning pipeline before execution

Prompt compiler with reasoning transparency and governance for controlled LLM execution.

HN linuxrebe1 22d ago

Turning a pile of documents into a searchable useable knowledge base

Tool to convert document collections into searchable knowledge base. LLM application for document processing.

HN johnfahey 22d ago

Show HN: Nully – FOSS AI chat without the bloat

Nully: Minimal open-source AI chat interface in Go with simple messaging and history, no advanced features.

HN lifeisstillgood 22d ago

Agentic test processes, LLM benchmarks, and other notes on agentic coding fr

Research notes on agentic coding processes, test automation approaches, and LLM benchmarks (insufficient content provided).

HN yr_animesh 22d ago

Show HN: Onboard-CLI, a LLM powered and AST-based tool to visualize codebase

Onboard-CLI: AST-based command-line tool using LLMs to visualize and map complex codebases via Tree-sitter parsing and React Flow canvas.

HN chand249 22d ago

Show HN: Skill-extractor turns coding agent transcripts into reusable skills

Skill-extractor tool extracts reusable skills from coding agent transcripts for automated task automation.