Isolater - Feed

HN pavpanchekha 4/9/2026

Can LLMs accelerate science? An experiment

Experimental analysis of whether LLMs can accelerate scientific research. Tests Opus and Codex on wet lab tasks.

HN paulpauper 4/9/2026

When AI Day of Reckoning?

Analysis of AI ROI expectations and failure rates. Discusses coding as LLMs' most promising application area.

HN 1270011 4/9/2026

I rebuilt Claude Code's removed /buddy companion as a permanent MCP app

Open-source MCP app restoring Claude's removed /buddy companion feature as standalone assistant compatible with any Claude Code version.

HN handfuloflight 4/9/2026

The Waymo Rule for AI-Generated Code

Article argues AI-generated code only needs to exceed developer capability, discusses formal methods and language design.

HN bobchen1322 4/9/2026

GPU solver that matches Fujitsu's $1M Digital Annealer on cubic optimization

GPU-based solver for higher-order binary optimization (HUBO) problems with 3+ variable interactions, matching commercial annealer performance.

Ax Cameron Pattison, Lorenzo Manuali, Seth Lazar 4/9/2026

Blind Refusal: Language Models Refuse to Help Users Evade Unjust, Absurd, and Illegitimate Rules

Research documenting patterns of language model refusal on unjust or absurd rules, arguing for better moral reasoning in AI safety training.

Ax Shoaib Sadiq Salehmohamed, Jinal Prashant Thakkar, Hansika Aredla, Shaik Mohammed Omar, Shalmali Ayachit 4/9/2026

Weakly Supervised Distillation of Hallucination Signals into Transformer Representations

Research on distilling hallucination detection signals into transformer representations during training, enabling inference-time detection without external verification.

Ax Isaac Henry, Avery Byrne, Christopher Giza, Ron Henry, Shahram Yazdani 4/9/2026

SymptomWise: A Deterministic Reasoning Layer for Reliable and Efficient AI Systems

Framework combining expert medical knowledge with deterministic reasoning to improve reliability and reduce hallucinations in AI-driven symptom analysis systems.

Ax Satwik Pandey, Suresh Raghu, Shashwat Pandey 4/9/2026

SELFDOUBT: Uncertainty Quantification for Reasoning LLMs via the Hedge-to-Verify Ratio

Research paper on uncertainty quantification for reasoning LLMs using hedge-to-verify ratio, addressing limitations of sampling and single-pass proxy methods for proprietary APIs.

Ax Varun Pratap Bhardwaj 4/9/2026

Qualixar OS: A Universal Operating System for AI Agent Orchestration

Application-layer OS for universal AI agent orchestration supporting 10 LLM providers, 8+ frameworks, 12 multi-agent topologies, and heterogeneous systems.

Ax Kranthi Kommuru, Kunal Khanvilkar, Gaurav Parekh 4/9/2026

ProofSketcher: Hybrid LLM + Lightweight Proof Checker for Reliable Math/Logic Reasoning

Hybrid LLM plus lightweight proof checker for reliable math/logic reasoning, verifying arguments and catching logical missteps in generated proofs.

Ax Jiaju Lin, Xingjian Du, Qingyun Wu, Ellen Wenting Zou, Jindong Wang 4/9/2026

On Emotion-Sensitive Decision Making of Small Language Model Agents

Studies emotion-sensitive decision-making in small language model agents using activation steering and game-theoretic evaluation.

Ax Qihan Ren, Peng Wang, Ruikun Cai, Shuai Shao, Dadi Guo, Yuejin Xie, Yafu Li, Quanshi Zhang, Xia Hu, Jing Shao, Dongrui Liu 4/9/2026

Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability

Analysis of cross-domain generalization in reasoning SFT with chain-of-thought, showing generalization is conditional on optimization, data, and base model capability.

Ax Monirul Islam Pavel, Siyi Hu, Muhammad Anwar Masum, Mahardhika Pratama, Ryszard Kowalczyk, Zehong Jimmy Cao 4/9/2026

KD-MARL: Resource-Aware Knowledge Distillation in Multi-Agent Reinforcement Learning

Knowledge distillation framework for multi-agent RL enabling resource-aware deployment on edge devices with smaller models.

Ax Yujun Cheng, Enfang Cui, Hao Qin, Zhiyuan Liang, Qi Xu 4/9/2026

AgentGate: A Lightweight Structured Routing Engine for the Internet of Agents

Lightweight routing engine for Internet of Agents managing agent discovery and request dispatch across devices, edge, and cloud with latency/privacy constraints.

Ax Samuel Sameer Tanguturi 4/9/2026

ATANT: An Evaluation Framework for AI Continuity

Open evaluation framework ATANT for measuring continuity in AI systems: persistence, context updating, and reconstruction across time.

Ax Jianhong Pang, Ruoxi Cheng, Ziyi Ye, Xingjun Ma, Zuxuan Wu, Xuanjing Huang, Yu-Gang Jiang 4/9/2026

Steering the Verifiability of Multimodal AI Hallucinations

Framework for steering verifiability of multimodal LLM hallucinations, distinguishing between obvious and elusive hallucinations to guide mitigation strategies.

Ax Juan Du, Yueteng Wu, Pan Zhao, Yuze Liu, Min Zhang, Xiaobin Xu, Xinglong Zhang 4/9/2026

TurboAgent: An LLM-Driven Autonomous Multi-Agent Framework for Turbomachinery Aerodynamic Design

LLM-driven autonomous multi-agent framework for end-to-end turbomachinery aerodynamic design, coordinating geometry, prediction, optimization, and validation.

Ax Shivanshu Shekhar, Sagnik Mukherjee, Jia Yi Zhang, Tong Zhang 4/9/2026

FVD: Inference-Time Alignment of Diffusion Models via Fleming-Viot Resampling

Inference-time alignment method for diffusion models using Fleming-Viot resampling to prevent diversity collapse in SMC sampling.

Ax Suhaas Garre, Erik Knutsen, Sushant Mehta, Edwin Chen 4/9/2026

Riemann-Bench: A Benchmark for Moonshot Mathematics

Benchmark for mathematical reasoning beyond competition math, testing advanced theoretical knowledge and deep mathematical reasoning.

Ax Zonghuan Xu, Xiang Zheng, Yutao Wu, Xingjun Ma 4/9/2026

Beyond Surface Judgments: Human-Grounded Risk Evaluation of LLM-Generated Disinformation

Evaluates LLM-generated disinformation risk by comparing LLM judges to human evaluation, addressing limitations of automated assessment.

Ax Daniele Fossem\`o, Filippo Mignosi, Giuseppe Placidi, Luca Raggioli, Matteo Spezialetti, Fabio Aurelio D'Asaro 4/9/2026

Explaining Neural Networks in Preference Learning: a Post-hoc Inductive Logic Programming Approach

Using inductive logic programming to approximate neural networks for user preference learning with explainability.

Ax Songze Li, Xiaoke Guo, Tianqi Liu, Biao Yi, Zhaoyan Gong, Zhiqiang Liu, Huajun Chen, Wen Zhang 4/9/2026

What's Missing in Screen-to-Action? Towards a UI-in-the-Loop Paradigm for Multimodal GUI Reasoning

GUI reasoning paradigm called UI-in-the-Loop for improved UI understanding and interaction, enhancing interpretability in screen-to-action tasks.

Ax Yunbo Long, Yunhan Liu, Liming Xu 4/9/2026

EmoMAS: Emotion-Aware Multi-Agent System for High-Stakes Edge-Deployable Negotiation with Bayesian Orchestration

Multi-agent system using small language models for emotion-aware negotiation with edge deployment focus, combining Bayesian orchestration and emotional dynamics.

Ax Irina Ar\'evalo, Marcos Oliva 4/9/2026

CAFP: A Post-Processing Framework for Group Fairness via Counterfactual Model Averaging

Post-processing framework for fairness in ML models via counterfactual averaging, applicable to deployed systems without full model control.

Ax Deliang Wen, Ke Sun, Yu Wang 4/9/2026

A-MBER: Affective Memory Benchmark for Emotion Recognition

Benchmark for evaluating emotion recognition in AI assistants over time, addresses long-term memory and emotional understanding in interactive systems.

Ax Alberto Pozanco, Marianela Morales, Pietro Totis, Daniel Borrajo 4/9/2026

Planning Task Shielding: Detecting and Repairing Flaws in Planning Tasks through Turning them Unsolvable

Research on detecting and repairing flaws in planning tasks by making them unsolvable, relevant to automated reasoning but not core AI agent/LLM focus.

Ax Jianfei Wu, Zhichun Wang, Zhensheng Wang, Zhiyu He 4/9/2026

EVGeoQA: Benchmarking LLMs on Dynamic, Multi-Objective Geo-Spatial Exploration

EVGeoQA benchmark evaluating LLMs on dynamic multi-objective geo-spatial exploration with user location constraints and compound reasoning beyond static retrieval.

Ax Yu Li, Sizhe Tang, Tian Lan 4/9/2026

Reason in Chains, Learn in Trees: Self-Rectification and Grafting for Multi-turn Agent Policy Optimization

T-STAR framework using tree-structured reinforcement learning with self-rectification and credit grafting to optimize multi-turn LLM agent policies under sparse rewards.

Ax Seongwoo Jeong, Seonil Son 4/9/2026

How Much LLM Does a Self-Revising Agent Actually Need?

Empirical study decomposing LLM-based agent competence to identify which capabilities derive from the language model versus explicit structural design in self-revising agents.

Ax Andrea Schimmenti, Valentina Pasqual, Fabio Vitali, Marieke van Erp 4/9/2026

Knowledge Graphs Generation from Cultural Heritage Texts: Combining LLMs and Ontological Engineering for Scholarly Debates

ATR4CH methodology for LLM-based knowledge graph extraction from cultural heritage documents using ontological engineering, validated on authenticity debate case study.

Ax Yonas Kassa, James Bonacci, Ping Wang 4/9/2026

Fighting AI with AI: AI-Agent Augmented DNS Blocking of LLM Services during Student Evaluations

Proposes AI-agent-augmented DNS blocking to prevent student access to LLM services during academic evaluations, addressing assessment integrity concerns.

Ax Nguyen Phuc Tran, Brigitte Jaumard, Oscar Delgado, Tristan Glatard, Karthikeyan Premkumar, Kun Ni 4/9/2026

LLM-Augmented Knowledge Base Construction For Root Cause Analysis

Evaluates LLM-augmented knowledge base construction for root cause analysis in network communications to enable rapid failure diagnosis and outage resolution.

Ax Yingjun Dai, Ahmed El-Roby 4/9/2026

EviSnap: Faithful Evidence-Cited Explanations for Cold-Start Cross-Domain Recommendation

EviSnap framework for cold-start cross-domain recommendation systems providing evidence-cited, auditable rationales through distilled review factorization without opaque embeddings.

Ax Kyubyung Chae, Jewon Yeom, Jeongjae Park, Seunghyun Bae, Ijun Jang, Hyunbin Jin, Jinkwan Jang, Taesup Kim 4/9/2026

Beyond Case Law: Evaluating Structure-Aware Retrieval and Safety in Statute-Centric Legal QA

Introduces SearchFireSafety benchmark for statute-centric legal QA addressing hierarchical document retrieval gaps and hallucination in regulatory reasoning with LLMs.

Ax Weishu Chen, Zhouhui Hou, Mingjie Zhan, Zhicheng Zhao, Fei Su 4/9/2026

Robustness Risk of Conversational Retrieval: Identifying and Mitigating Noise Sensitivity in Qwen3-Embedding Model

Empirical study of embedding-based retrieval robustness in conversational settings, identifying vulnerability in Qwen3-embedding models to structured dialogue-style noise.

Ax Yuelin Hu, Zhengxue Cheng, Ronghua Wu, Qunshan Gu, Hongwei Hu, Wei Liu, Qiao Liang, Li Song 4/9/2026

WebExpert: domain-aware web agents with critic-guided expert experience for high-precision search

Domain-aware web agent with critic-guided experience retrieval and schema-light facet induction for high-precision search in finance, biomedicine, and pharmaceutical domains.

Ax Yichen Gong, Zhuohan Cai, Sunhao Dai, Yuqi Zhou, Zhangxuan Gu, Changhua Meng, Shuheng Shen 4/9/2026

VenusBench-Mobile: A Challenging and User-Centric Benchmark for Mobile GUI Agents with Capability Diagnostics

Online benchmark VenusBench-Mobile for evaluating mobile GUI agents under realistic user-centric conditions with capability diagnostics across diverse app-agnostic tasks.

Ax Peijie Yu, Wei Liu, Yifan Yang, Jinjian Li, Zelong Zhang, Xiao Feng, Feng Zhang 4/9/2026

Benchmarking LLM Tool-Use in the Wild

Benchmark for evaluating LLM tool-use agents on multi-turn, multi-step interactions addressing compositional tasks, implicit intent, and instruction transitions in real user behavior.

Ax Peter Kirgis, Ben Hawriluk, Sherrie Feng, Aslan Bilimer, Sam Paech, Zeynep Tufekci 4/9/2026

LLM Spirals of Delusion: A Benchmarking Audit Study of AI Chatbot Interfaces

Benchmarking study measuring how different LLMs resist or escalate delusional and conspiratorial thinking in sustained open-ended conversations.

Ax Mar Gonz\`alez I Catal\`a, Haitz S\'aez de Oc\'ariz Borde, George D. Monta\~nez, Pietro Li\`o 4/9/2026

The Stepwise Informativeness Assumption: Why are Entropy Dynamics and Reasoning Correlated in LLMs?

Investigates correlation between internal entropy dynamics and reasoning correctness in autoregressive LLMs, proposing stepwise informativeness assumption to explain phenomenon.

Ax Angelina Hintsanen 4/9/2026

Hallucination as output-boundary misclassification: a composite abstention architecture for language models

Frames LLM hallucinations as output-boundary misclassification and proposes composite intervention combining instruction-based refusal with structural abstention gate using support deficit scoring.

Ax Tianyi Huang, Ming Hou, Jiaheng Su, Yutong Zhang, Ziling Zhang 4/9/2026

Consistency-Guided Decoding with Proof-Driven Disambiguation for Three-Way Logical Question Answering

Consistency-guided decoding approach for three-way logical QA in LLMs, addressing negation inconsistency and epistemic Unknown failures.

Ax Sayantan Kumar, Jeremy C. Weiss 4/9/2026

Temporally Phenotyping GLP-1RA Case Reports with Large Language Models: A Textual Time Series Corpus and Risk Modeling

Textual time-series corpus of 136 GLP-1RA case reports with LLM-based timeline extraction and risk modeling for longitudinal analysis.

Ax Danbo Chen, Zijun Zhou, Yongyang Cai, Jiahong Qin, Ani Katchova, Lei Chen 4/9/2026

Concentrated siting of AI data centers drives regional power-system stress under rising global compute demand

Framework combining LLM analysis with energy-system modeling to forecast electricity footprint of AI data centers through 2030.

Ax Pei-Fu Guo, Ya-An Tsai, Chun-Chia Hsu, Kai-Xin Chen, Yun-Da Tsai, Kai-Wei Chang, Nanyun Peng, Mi-Yen Yeh, Shou-De Lin 4/9/2026

Beyond Facts: Benchmarking Distributional Reading Comprehension in Large Language Models

Text2DistBench benchmark for evaluating LLM ability to infer distributional knowledge and population-level trends from text collections.

Ax O. Ibrahimzade, K. Tabasaransky 4/9/2026

Cross-Lingual Transfer and Parameter-Efficient Adaptation in the Turkic Language Family: A Theoretical Framework for Low-Resource Language Models

Theoretical framework for cross-lingual transfer and parameter-efficient adaptation in low-resource Turkic languages using LLMs.

Ax Hansoo Lee, Rafael A. Calvo 4/9/2026