All Updates

Page 1191 of 1475

March 11, 2026

โš›๏ธ
้‡ๅญไฝโ€ข114d ago

2B Funding, 7x Valuation Surge for Young Robotics Firm

A robotics company founded by post-00s entrepreneurs received 2 billion RMB in funding. Its valuation exploded 7 times in one year, backed by national-level capital. Data acquisition costs are just 10% of traditional schemes.

#robotics-funding#data-acquisition#startup-valuation
๐Ÿ’ฐ
้’›ๅช’ไฝ“โ€ข114d ago

Lobster Sparks Global AI Token Migration

A major token migration, termed global 'west data east compute,' was triggered by 'Lobster.' This positions it as the AI era's 'world Token factory.' It highlights shifts in AI compute infrastructure.

#token-migration#data-compute#ai-factory
๐Ÿ’ฐ
้’›ๅช’ไฝ“โ€ข114d ago

Global LLMs Collapse on Chinese Wenyanwen

Leading global large language models suffer total defeat against Chinese classical Chinese (wenyanwen). This deficiency underscores that wenyanwen is far from the only underlying vulnerability in LLMs.

#wenyanwen#benchmark-failure#multilingual
๐Ÿฏ
Huxiu (่™Žๅ—…)โ€ข114d ago

Tanji Tech Gains Control of Zhen'ai Meijia

Tanji Technology has completed share transfer, becoming the controlling shareholder of listed firm Zhen'ai Meijia with 29.99% stake. They plan a partial tender offer for 15% more shares, totaling 44.99%. As an AI unicorn, Tanji specializes in digital employees via its Taiqing platform for sales and marketing agents.

#ai-agents#acquisition#digital-employees
๐Ÿ“„
ArXiv AIโ€ข114d ago

Time, Identity & Consciousness in LM Agents

This arXiv paper introduces a method to evaluate machine consciousness in language model agents by applying Stack Theory's temporal gap to separate behavior ingredients from co-instantiation. It instantiates Arpeggio and Chord postulates for two persistence scores from scaffold traces. The toolkit maps scaffolds into an identity morphospace, revealing tradeoffs and connecting to operational identity metrics.

#lm-agents#identity-metrics
๐Ÿ’ผ
VentureBeatโ€ข114d ago

RSAC 2026 Sandbox Unveils AI Security Finalists

RSAC Innovation Sandbox has generated $18.1B in alumni investments and over 100 acquisitions, fostering a cycle of founder-led innovation in cybersecurity. Alumni successes include Securiti AI's $2.7B sale to Veeam and multiple CrowdStrike acquisitions. 2026 top 10 finalists tackle AI governance, agentic AI security, and social engineering via tools like Charm Security and Geordie AI.

#cybersecurity#ai-security#startups
๐Ÿ“„
ArXiv AIโ€ข114d ago

MEMO Boosts LLM Win Rates in Multi-Agent Games

MEMO is a self-play framework optimizing inference-time context for multi-turn multi-agent LLM games, addressing variance and underperformance. It couples memory retention from trajectories with prompt evolution using TrueSkill and prioritized replay. Achieves 25% to 49.5% win rate gains for GPT-4o-mini across five text games.

#multi-agent#self-play#context-optimization
๐Ÿ“„
ArXiv AIโ€ข114d ago

Meissa: Lightweight Offline Medical AI Agent

Meissa is a 4B-parameter multi-modal medical LLM that enables offline agentic capabilities by distilling trajectories from frontier models like GPT. It features unified trajectory modeling, three-tier stratified supervision for strategy selection, and prospective-retrospective learning for execution. Trained on 40K trajectories, it matches or exceeds proprietary agents across 13 medical benchmarks with 25x fewer parameters and 22x lower latency; fully open-sourced.

#medical-ai#agentic-llm#multi-modal
๐Ÿ“„
ArXiv AIโ€ข114d ago

MASEval: System-Level Multi-Agent Evaluation

MASEval introduces a framework-agnostic library for evaluating entire LLM-based multi-agent systems, beyond just models. It assesses impacts of topology, orchestration, and error handling across benchmarks. Results show framework choice rivals model choice in performance.

#multi-agent#benchmarks#agentic-systems
๐Ÿ“„
ArXiv AIโ€ข114d ago

LDP: Identity-Aware Multi-Agent LLM Protocol

LDP is a new AI-native protocol for multi-agent LLM systems, featuring delegate identity cards, progressive payloads, governed sessions, provenance tracking, and trust domains. Implemented as a JamJet plugin, it reduces latency by 12x on easy tasks, cuts token usage by 37%, and improves security in simulations. Evaluations highlight efficiency gains but note limitations in small delegate pools and noisy provenance.

#multi-agent#delegation-protocol#llm-efficiency
๐Ÿ“„
ArXiv AIโ€ข114d ago

Guardian: Multi-LLM Consensus for Missing Persons

Guardian is an end-to-end system supporting missing-child investigations with a multi-LLM pipeline for information extraction and search planning. It coordinates task-specialized LLMs and uses a consensus engine to resolve output disagreements. The pipeline employs QLoRA fine-tuning on curated datasets for conservative, auditable performance.

#consensus-engine#missing-persons#qlora-fine-tuning
๐Ÿ“„
ArXiv AIโ€ข114d ago

Guardian AI for Missing-Child Search Planning

Guardian is an end-to-end decision-support system that processes unstructured missing-child case data into spatiotemporal representations and generates probabilistic search plans over 0-72 hours. It features a three-layer predictive model: interpretable Markov chains for risk surfaces, reinforcement learning for search optimization, and LLMs for quality assurance. Synthetic case studies demonstrate effective, interpretable outputs for zone prioritization.

#missing-child-ai#rl-llm-hybrid
๐Ÿ“„
ArXiv AIโ€ข114d ago

FABRIC Verifies Neural Feedback Systems

Researchers introduce FaBRIC, integrating forward and backward reachability analysis for verifying neural feedback systems. It provides scalable over- and underapproximations of backward reachable sets. Evaluations on benchmarks show it outperforms prior state-of-the-art methods.

#neural-feedback#verification
๐Ÿ“„
ArXiv AIโ€ข114d ago

EPOCH: Agentic Protocol for Multi-Round Optimization

EPOCH introduces a unified engineering protocol for multi-round system optimization using autonomous agents. It divides processes into baseline construction and iterative self-improvement phases, with role-constrained stages for planning, implementation, and evaluation. The protocol standardizes execution across prompts, code, models, and rules, ensuring stability, reproducibility, and traceability.

#autonomous-agents#self-improvement
๐Ÿ“„
ArXiv AIโ€ข114d ago

Budget-Constrained Agentic LLM Search Study

This arXiv paper presents a measurement study on how search depth, retrieval strategy, and completion budget impact accuracy and cost in agentic RAG systems under fixed constraints. Using the BCAS evaluation harness, it compares six LLMs across three QA benchmarks, finding hybrid retrieval with re-ranking yields top gains and larger budgets aid synthesis tasks.

#agentic-search#retrieval-strategy#budget-optimization
๐Ÿ“„
ArXiv AIโ€ข114d ago

AI Agent Masters RPM Triage in Minutes

Sentinel, an autonomous AI agent, uses Model Context Protocol (MCP) for reliable triage of remote patient monitoring vitals, outperforming individual clinicians in emergency sensitivity (97.5% vs 60%). It achieves 95.8% emergency sensitivity against human consensus at $0.34 per triage. The agent enables scalable monitoring to reduce mortality like TIM-HF2 without high costs.

#healthcare-ai#autonomous-agents#clinical-triage
๐Ÿ“„
ArXiv AIโ€ข114d ago

AgentOS: Natural Language OS Paradigm

Proposes AgentOS, a Personal Agent Operating System that replaces GUI desktops with a natural language interface. An Agent Kernel interprets user intent, coordinates agents, and turns apps into modular Skills composed via natural language rules. Frames OS as a KDD pipeline for real-time knowledge discovery and workflow automation.

#nui#agent-kernel#skills-modules
๐Ÿฏ
่™Žๅ—…โ€ข114d ago

Blind Patients Regain Vision via Retinal Chip

Science Corp's PRIMA device restored structured vision in 80% of 38 AMD-blind patients, enabling letter reading and novel completion, per NEJM study. Max Hodak outlines BCI roadmap from neural engineering to biological interfaces and consciousness longevity. He predicts first 1000-year humans already born by 2035.

#bci#neural-interface#longevity
๐Ÿ“Š
Bloomberg Technologyโ€ข114d ago

China Curbs OpenClaw AI in Banks, Agencies

Chinese authorities are restricting state-run enterprises and government agencies from running OpenClaw AI apps on office computers. This swift action addresses potential security risks amid widespread experimentation with agentic AI by companies and consumers across China.

#china-regulation#agentic-ai#security-risk
๐Ÿ’ฐ
้’›ๅช’ไฝ“โ€ข114d ago

Viral Crayfish Pet Eyes Hotel Industry

Online viral crayfish raising trend questions entry into hotel business. Future favors AI experts, positioning crayfish as mere prelude to bigger things.

#viral-app#china-consumer#ai-business
Page 1191 of 1475