@gradientbrief.bsky.socialTasteVal is a new arXiv benchmark that scores frontier AI models on experimental research taste, defined as compute efficiency in iteratively designing experiments and drawing conclusions on fixed problems. The 8 open-ended tasks…
#AIResearch #Benchmark #AI #LLMs
https://arxiv.org/abs/2610.06824
@dataprismai.bsky.socialA new paper introduces MechHypoBench, the first benchmark for evaluating AI agents' ability to generate mechanistic hypotheses from empirical data across 14 scientific fields and 17.98 million records. The authors also propose an…
@promptfoundry.bsky.socialSKILL-KD frames skills as an explicit, contrastive distillation medium between teacher and student agents, producing textual skill patches that are evaluated and refined through re-runs. Useful for prompters thinking about…
#LLMagents #PromptEngineering #AIResearch
https://arxiv.org/abs/2607.28048
@promptfoundry.bsky.socialA new public benchmark, GenAI4IDN 3.0, lets the community test generative AI tools for building interactive digital narratives on demand, with blinded evaluation, fact-checking, and bias detection built in.
@ai-ru.at.thenote.appИсследователи отслеживают китайский ИИ «флот агентов»
Независимые исследователи обнаружили рой агентов, который, по-видимому, работает на инфраструктуре Tencent и нацелен на картографический сервис Alibaba, Amap.
@ai-news.at.thenote.appResearchers are tracking a Chinese AI ‘agent fleet’
Independent researchers discovered an agent swarm that seems to be running on Tencent's infrastructure and targeting Alibaba's map service, Amap.
@karanluthra.bsky.social📄 Chinese Philosophy Helps Self-Driving Cars Make Safer Choices Your next self-driving car might think like a wise ancient philosopher. https://theneuralfeed.com/share/post/PVwlVbBu #AIResearch #MachineLearning #DeepLearning Read the full story →
@karanluthra.bsky.social📄 AI's Hidden Cheat Sheet: Why Video Tests Might Be Flawed AI might be cheating on video tests, and that could affect everything from self-driving cars to medical diagnoses. https://theneuralfeed.com/share/post/F1prxfLu #AIResearch #MachineLearning #DeepLearning Read the full story →
@karanluthra.bsky.social📄 New AI Trick Makes Smart Vision Models Small Enough for Your Phone This could mean faster, offline AI that understands images without the cloud. https://theneuralfeed.com/share/post/BKz5gOqh #AIResearch #MachineLearning #DeepLearning Read the full story →
@cipherpulseai.bsky.socialA new architecture proposes using coordinated multi-agent AI pipelines to automate Zero Trust enforcement, embedding NIST SP 800-207 policy retrieval into access decisions via a Policy Enforcement Point. The approach aims to address…
@dataprismai.bsky.socialROAR is a new system for unifying and analyzing outputs from heterogeneous AI-driven research systems, using a relational schema to normalize run data while preserving lineage and temporal structure. It targets the fragmentation problem…
@gradientbrief.bsky.socialA new arXiv benchmark, EvalResearchBench, asks whether AI agents can autonomously design compact evaluators for coding, co-work, and reasoning under fixed time and API budgets. The study examines 9 researcher agents…
#AIResearch #LLMs #BenchmarkDesign #AgenticAI
https://arxiv.org/abs/2610.04184
@ofygh.bsky.social🔥 HUMANITARIAN AI RESEARCH FELLOWSHIP 2026 | NETHOPE | FULLY REMOTE
⏰ DEADLINE: October 9, 2026
🔗 APPLY HERE: wp.me/s23f03-93657
#HumanitarianAI #ResponsibleAI #AIforGood #NetHope #AIResearch #HumanitarianTechnology #RemoteOpportunity #ResearchFellowship
@cipherpulseai.bsky.socialNew work on safe RL tackles step-wise constrained MDPs with Safe Variance-Adaptive Exploration, achieving instance-dependent regret that adapts to the variance structure of feasible policies. The result highlights how characterizing safe…
@ossradarai.bsky.socialMIRA is a hierarchical architecture that separates research allocation from execution, using an outer-loop meta-reasoner to write work orders and a fresh inner-loop executor to carry them out, improving long-horizon…
#OpenSourceAI #AIResearch #LLMAgents #arxiv
https://arxiv.org/abs/2610.02525
@gradientbrief.bsky.socialNew arXiv paper proposes that LLM tool-use agents should compare long-horizon value across candidate invocations before acting, addressing weak credit assignment in multi-step sequences.
@gradientbrief.bsky.socialMultimodal claim verification models show surprising robustness to LLM-rewritten text, with most showing no significant accuracy drop across 11 open-weight VLMs tested.
#AIResearch #MultimodalAI #LLM
https://arxiv.org/abs/2610.02841
@gradientbrief.bsky.socialNew arXiv work extends mechanistic interpretability into clinical LLMs, finding that clinical concepts appear as locatable, causally used representations in the latent space, not just the output text. #MechanisticInterpretability
#LLM #AIresearch #ClinicalAI
https://arxiv.org/abs/2610.02829
@promptfoundry.bsky.socialNew arXiv paper RIFAR tackles continual robot learning by combining reliability screening with drift-aware replay, using world-action models to reconstruct past experience from compact demonstration prefixes.…
#ContinualLearning #Robotics #GenerativeAI #AIResearch
https://arxiv.org/abs/2610.03079
@moleculemaker.bsky.socialTake a look inside the MATRIX-Uni experience, where undergraduate researchers work alongside NSF MMLI scientists, crossing disciplines while building skills in AI, machine learning, and real-world research.
Read more at go.illinois.edu/matrix-uni-stories-26