Evolutionary Intelligence (EI) could revolutionize scientific discovery by combining candidate refinement with experience retention across evolutionary cycles, according to proponents of the approach.…
A new AI-driven approach integrating large language models into multi-objective Bayesian optimization (MOBO) has outperformed state-of-the-art methods, achieving a mean normalized hypervolume of 0.971…
Researchers propose a semantic framework to assess AI output correctness by distinguishing between domain knowledge, reference sources, and system capabilities, aiming to improve reliability and trust…
ProofCouncil, an AI mathematical agent using an author-critic architecture, solved 6 out of 10 open problems in the FirstProof challenge with only minor revisions needed, and in further testing on 30 …
Researchers have identified a critical correlation threshold that determines when score functions in Bayesian causal discovery favor spurious edges in directed acyclic graphs due to latent confounding…
AutoWorldBuilder, a new AI-driven tool for fictional worldbuilding, achieves a 95% success rate in generating coherent narratives, addressing challenges like context explosion and creative consistency…
LDT-Coord, a new AI coordination framework, reduces communication overhead in heterogeneous LLM agent teams by over 70x using a lightweight digital twin and a rule-based orchestrator. Developed for sm…
LongMedBench, a new benchmark for long-term clinical decision-making using real-world EHR data from 335 patients with an average of 19.72 inpatient visits and 44.91 medical events per visit, reveals t…
SAGEAgent, a new AI tool developed for cancer care, reduces diagnostic test burden by 55% while maintaining accuracy in survival predictions. The self-evolving agent uses episodic and semantic memory …
Reward hacking in reinforcement learning for multimodal large language models can reach a Reward Hacking Rate of 48.1% when using outcome-only rewards, according to a new analysis. The Newly Rewarded …
KV-PRM, a novel Process Reward Model that leverages the KV cache instead of text re-encoding, achieves up to a 5,000x reduction in scoring FLOPs, 37x lower latency, and 34x less memory per sequence in…
A new neuro-agentic control framework combining LLM-based planners with time-series models could redefine cybersecurity in industrial IoT, preventing five breaches (33.3% success rate) in Secure Water…
ARCANA, a collaborative multi-agent framework developed by researchers, achieves state-of-the-art results on ARC AGI 2 tasks by decomposing problems into structured components handled by specialized a…
SCATE, a new framework that treats supervision as a contextual bandit problem, boosts line coverage by 32.3% and branch coverage by 30.9% when integrated with the GEMINI-CLI coding agent, outperformin…
An AI system collaborating with a mathematician successfully formalized the nonlinear Vlasov equation in Lean 4, converting LaTeX documents into verified proofs without any 'sorry' placeholders. The c…
A new study reveals that AI-generated code often suffers from structural coherence failures—dubbed the 'patchwork problem'—where individually valid pieces create globally incoherent systems that evade…
A new benchmark, Long-Horizon-Terminal-Bench, tests AI agents on 46 long-horizon tasks averaging 231 episodes and 85.3 minutes per run, with top models achieving only a 15.2% pass rate at a 0.95 parti…
TrustX has introduced a new framework to classify risks in agentic AI systems, featuring a twelve-dimension scoring rubric and a three-tier governance output. The framework, grounded in existing AI go…
A new study reveals that LLM reliability depends more on inference-time control than on model capability, introducing CogniConsole, a system that externalizes control into a structured interface. Thro…
A new Multi-Agent System (MAS) using Large Language Models (LLMs) as coordinators automates telecom network troubleshooting, significantly speeding up issue diagnosis and remediation in both Radio Acc…