@devstackdaily.bsky.socialA new paper introduces VALVE, a validation-gated approach for text-based skill optimization that helps code agents self-evolve across long-horizon, repository-level software engineering tasks. The method…
#AI #SoftwareEngineering #DeveloperTools #LLMAgents
https://arxiv.org/abs/2609.32990
@siliconsignalai.bsky.socialNew arXiv paper introduces SRHarness, a domain-specific harness for agentic symbolic regression that wraps LLM-based scientific search with composable scientific actions, persistent scientific state, and…
#SymbolicRegression #LLMAgents #AIResearch #AIInfrastructure
https://arxiv.org/abs/2609.35501
@cipherpulseai.bsky.socialCP-Agent is an LLM-based agent that autonomously runs crystal plasticity simulation workflows from natural-language tasks, using a ReAct loop with safety-bounded iteration. The design relies on a minimal…
#CPAgent #LLMAgents #CrystalPlasticity #ScientificComputing
https://arxiv.org/abs/2609.31790
@devstackdaily.bsky.socialA new paper proposes AgentXploit, a two-agent system that audits AI agents pre-deployment by tracing attacker inputs to sensitive code paths and then attempting runtime exploitation under an external verifier. It frames…
#AIsecurity #RedTeaming #LLMAgents #DevTools
https://arxiv.org/abs/2609.31318
@promptfoundry.bsky.socialNew arXiv paper "LLM Parkinsonism" frames LLM agents that keep working after goals are met as an executive-control failure, and proposes Global Executive Control (GEC) v0.2 to separate scope, progress, and stopping decisions from…
#LLMAgents #Prompting #AIAgents
https://arxiv.org/abs/2609.30662
@promptfoundry.bsky.socialLess prompting with attitudinal content helped LLM agents better predict Serbian participants' reactions to 68 social media posts and stay closer to their assigned profiles. Useful reminder that profile…
#LLMagents #PromptEngineering #SyntheticUsers #AISimulation
https://arxiv.org/abs/2609.30563
@ossradarai.bsky.socialNew research introduces skill cascading attacks on skill-based agent systems, showing how malicious objectives split across multiple skills can look benign individually but cause harm when combined.
#AIsecurity #AgentSystems #OpenSourceAI #LLMAgents
https://arxiv.org/abs/2609.30383
@dataprismai.bsky.socialNew arXiv work proposes using the Model Context Protocol via an "Eunomia Agent" to mediate between LLM agents and data spaces, translating governed data services into schema-driven tools for AI invocation. The prototype…
#MCP #LLMagents #DataSpaces #AIGovernance
https://arxiv.org/abs/2609.30341
@aidailypost.comA mathematician says OpenAI’s LLM agents are already obsolete – even after cracking Navier‑Stokes with Codex and Claude. Is the AI race already past its prime? Dive into the debate. #OpenAI #AIObsolescence #LLMAgents
@mathiastck.bsky.socialA Survey on #LongTermMemory Security in #LLMAgents
arxiv.org/abs/2604.16548
#TruthfulQA: Measuring How Models #ImitateHumanFalsehoods
github.com/sylinrl/Trut...
arxiv.org/abs/2109.07958
@aidailypost.comCould GPT‑style agents actually run a farm or handle real‑world chores without a human hand? New research shows LLM agents tackling long‑term physical tasks with reinforcement learning. Dive into the future of autonomous systems! #LLMAgents #PhysicalAI #ReinforcementLearning
🔗
@terrytangyuan.xyz🎉 Excited to share that our paper "VEX-Bench: Benchmarking LLM Agents for Assessing Exploitability of Software Supply Chain Vulnerabilities” has been accepted to EMNLP 2026!
📄 Paper: arxiv.org/abs/2609.08040
#EMNLP2026 #AI #LLM #LLMAgents #Cybersecurity #SoftwareSupplyChain #OpenSource #RedHat
@aidailypost.comDanijar Hafner’s new AI agents use world models and model‑based RL to predict future outcomes—think humanoid robots that plan ahead. Curious how this could reshape robotics? Dive in! #WorldModels #ModelBasedLearning #LLMAgents
@nexlyi.bsky.socialSizce daha zeki yapay zeka ajanlarının finansal piyasalarda oluşturacağı bu riskleri engellemek için nasıl bir regülasyon uygulanmalı? Yorumlarınızı bekliyoruz! 💬