惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
Visual Studio Blog
量子位
大猫的无限游戏
大猫的无限游戏
Hugging Face - Blog
Hugging Face - Blog
S
SegmentFault 最新的问题
Blog — PlanetScale
Blog — PlanetScale
月光博客
月光博客
Google DeepMind News
Google DeepMind News
小众软件
小众软件
WordPress大学
WordPress大学
宝玉的分享
宝玉的分享
MongoDB | Blog
MongoDB | Blog
B
Blog RSS Feed
博客园 - Franky
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
B
Blog
博客园 - 聂微东
The GitHub Blog
The GitHub Blog
Recent Announcements
Recent Announcements
Y
Y Combinator Blog
Microsoft Security Blog
Microsoft Security Blog
雷峰网
雷峰网
Jina AI
Jina AI
酷 壳 – CoolShell
酷 壳 – CoolShell

cs.SE updates on arXiv.org

VLA Foundry: A Unified Framework for Training Vision-Language-Action Models Evaluating LLM-Generated Obfuscated XSS Payloads for Machine Learning-Based Detection Do Agents Dream of Root Shells? Partial-Credit Evaluation of LLM Agents in Capture the Flag Challenges Refute-or-Promote: An Adversarial Stage-Gated Multi-Agent Review Methodology for High-Precision LLM-Assisted Defect Discovery From Particles to Perils: SVGD-Based Hazardous Scenario Generation for Autonomous Driving Systems Testing Choose Your Own Adventure: Non-Linear AI-Assisted Programming with EvoGraph Human-Machine Co-Boosted Bug Report Identification with Mutualistic Neural Active Learning LLMSniffer: Detecting LLM-Generated Code via GraphCodeBERT and Supervised Contrastive Learning Neurosymbolic Repo-level Code Localization CodeMMR: Bridging Natural Language, Code, and Image for Unified Retrieval Symbolic Guardrails for Domain-Specific Agents: Stronger Safety and Security Guarantees Without Sacrificing Utility Verification Modulo Tested Library Contracts The Semi-Executable Stack: Agentic Software Engineering and the Expanding Scope of SE Scaling Test-Time Compute for Agentic Coding AI-Assisted Requirements Engineering: An Empirical Evaluation Relative to Expert Judgment From Procedural Skills to Strategy Genes: Towards Experience-Driven Test-Time Evolution Atropos: Improving Cost-Benefit Trade-off of LLM-based Agents under Self-Consistency with Early Termination and Model Hotswap Vibe-Coding: Feedback-Based Automated Verification with no Human Code Inspection, a Feasibility Study Benchmarks for Trajectory Safety Evaluation and Diagnosis in OpenClaw and Codex: ATBench-Claw and ATBench-Codex Bounded Autonomy for Enterprise AI: Typed Action Contracts and Consumer-Side Execution AIPC: Agent-Based Automation for AI Model Deployment with Qualcomm AI Runtime Analyzing Chain of Thought (CoT) Approaches in Control Flow Code Deobfuscation Tasks Asking What Matters: Reward-Driven Clarification for Software Engineering Tasks Prompt-Driven Code Summarization: A Systematic Literature Review LinuxArena: A Control Setting for AI Agents in Live Production Software Environments LLMs taking shortcuts in test generation: A study with SAP HANA and LevelDB Large Language Models to Enhance Business Process Modeling: Past, Present, and Future Trends CollabCoder: Plan-Code Co-Evolution via Collaborative Decision-Making for Efficient Code Generation Sentiment analysis for software engineering: How far can zero-shot learning (ZSL) go? Learning from Change: Predictive Models for Incident Prevention in a Regulated IT Environment
Benchmarking Mythos-Linked Bug Rediscovery
Isaac David, Arthur Gervais · 2026-05-17 · via cs.SE updates on arXiv.org

Anthropic's April 2026 Mythos materials combine benchmark claims with concrete bug-finding stories across OpenBSD, FreeBSD, Linux, FFmpeg, and browsers. This paper reports a controlled target-file rediscovery experiment on six public or high-confidence Mythos-linked systems tasks. Each model receives the same target file or files, read-only source tools, three repeats per task, and one manual target-matching rubric; prompts omit CVE identifiers, patch hashes, advisory text, author names, disclosure dates, and answer key root cause language. The experiment contains 54 counted model-task attempts: three models, six tasks, and three repeats, giving 18 attempts per model. GPT-5.5 xhigh achieves 5/18 target rediscoveries, covering 2/6 tasks; counting one wrong-target mpegts.c finding separately gives 3/6 distinct core bugs. Claude Opus 4.7 achieves 1/18 target rediscoveries, covering 1/6 tasks. Kimi K2 records 0/18 target rediscoveries. The dominant failure mode is early commitment to plausible alternate candidates within the assigned file: models often submit source-grounded hypotheses while missing the specific invariant corrected by public Mythos patch evidence. These results do not refute Anthropic's undisclosed workflow, but show that under this favorable target-file scaffold, systems-specific prompting yields only six target matches across 54 counted attempts.