All Updates

Page 558 of 1667

May 20, 2026

🔥
36氪71d ago

Nvidia to Establish New R&D Center in Singapore

Nvidia is expanding its AI footprint by establishing a new R&D center in Singapore. This facility marks the company's first R&D hub in the country and its second in the Asia-Pacific region.

#ai-infrastructure#apac-market#semiconductor
🇨🇳
cnBeta (Full RSS)71d ago

DeepMind CEO: Replacing Developers with AI is a Mistake

DeepMind CEO Demis Hassabis argues that companies should focus on using AI to boost productivity rather than using it as a tool for layoffs. He emphasizes that AI's primary value lies in augmenting human capabilities.

#ai-strategy#workforce-management#productivity
🔥
36氪71d ago

Circular Intelligence Launches iMLite AI Edge Engine

Circular Intelligence has released iMLite AI, an edge-based real-time decision engine designed for outdoor smart hardware. It supports 100% offline operation and low-power hardware adaptation.

#edge-ai#iot#offline-first
🇨🇳
cnBeta (Full RSS)71d ago

SoftBank Internal Concerns Over Massive OpenAI Investment

SoftBank insiders are expressing concern over Masayoshi Son's massive $60 billion investment commitment to OpenAI. There is internal friction regarding the level of trust placed in Sam Altman.

#ai-funding#corporate-governance#venture-capital
🔥
36氪71d ago

Fliggy Launches AI Assistant for Hotel Merchants

Fliggy has introduced an AI-powered assistant for hotel merchants, allowing them to manage room availability, inventory, and invoices via natural language commands. It also provides data-driven pricing and service optimization strategies.

#llm-application#b2b-saas#automation
🐯
虎嗅71d ago

Chinese STEM talent returning from US

Analysis of the shifting landscape for Chinese students in the US, noting a trend of high-tech talent returning to China due to policy changes and economic factors.

#talent-mobility#geopolitics#hiring
🏠
IT之家71d ago

Sony Launches 10th Anniversary WH-1000X Edition

Sony has released the 1000X THE COLLEXION, a 10th-anniversary edition of its noise-canceling headphones featuring premium materials and the new V3 integrated processor.

#audio-ai#edge-computing#hardware-innovation
🐯
虎嗅71d ago

The Health Risks of Campus Light Meal Trends

The light meal market in Chinese universities is growing rapidly, but experts warn that improper 'clean eating' can lead to metabolic issues and nutrient deficiencies. The article highlights the importance of balanced nutrition over simple calorie counting.

#health-tech#consumer-behavior#nutrition-ai
🐯
虎嗅71d ago

Global Surge in Tea Ware Consumption Trends

Driven by social media aesthetics and a desire for 'slow living,' Western consumers are increasingly investing in traditional tea ware. This trend highlights the intersection of cultural appreciation and digital-age consumer behavior.

#e-commerce#consumer-trends#aesthetic-economy
🔥
36氪71d ago

Digua Robot and Qianxun Intelligence Partner on VLA

Digua Robot and Qianxun Intelligence have partnered to integrate the Spirit v1.5 VLA model with the Sunray S600 chip. This collaboration aims to provide a high-performance, mass-producible hardware-software solution for robotics.

#robotics#vla#edge-ai
📱
Engadget71d ago

Kickstarter reverses mature content policy after creator backlash

Kickstarter has decided to revert to its previous mature content guidelines following significant criticism from its creator community. The platform had introduced new rules last week that were met with immediate pushback.

#content-moderation#platform-policy#community-management
📊
Bloomberg Technology71d ago

Cathie Wood: Tesla Robotaxis Set for Massive Growth

Cathie Wood of Ark Investment Management identifies transportation as the primary revenue-generating opportunity for embodied AI. She predicts that Tesla's robotaxi initiatives are on the verge of significant market expansion.

#embodied-ai#autonomous-driving#transportation
📄
ArXiv AI71d ago

Scaling Document AI: Microservice Patterns for Production Pipelines

This research presents a robust microservice architecture for deploying document understanding pipelines at scale. It offers insights into optimizing OCR and LLM-based extraction for high-throughput production environments.

#document-ai#microservices#latency-optimization
📄
ArXiv AI71d ago

ReElicit: Optimizing System Prompts via Dynamic Bayesian Representations

ReElicit is a new Bayesian optimization framework that uses LLMs to elicit interpretable feature spaces for tuning system prompts. It enables effective prompt optimization even when only aggregate scalar feedback is available, outperforming traditional baselines.

#prompt-engineering#llm-tuning
📄
ArXiv AI71d ago

Provably Convergent Workflow Learning for Decentralized Multi-Agent LLMs

Researchers introduced IC-Q, a decentralized Q-learning algorithm that enables multi-agent LLM pipelines to coordinate via shared artifacts without centralized trajectory access. The study provides the first finite-sample convergence guarantee for neural Q-learning under decentralized partial observability.

#multi-agent-systems#llm-pipelines#decentralized-ai
📄
ArXiv AI71d ago

POLAR-Bench: Evaluating Privacy-Utility Trade-offs in LLM Agents

POLAR-Bench is a new diagnostic benchmark designed to test how LLM agents handle private user data when interacting with adversarial third-party systems. It reveals a significant performance gap in privacy alignment between frontier models and smaller open-weight models.

#llm-security#privacy-alignment#benchmarking
📄
ArXiv AI71d ago

New Framework Solves Interference in Multi-Task Machine Unlearning

Researchers introduced an interference-aware framework for multi-task unlearning that prevents unintended performance degradation when removing data. By using task-aware gradient projection and instance-level orthogonalization, the method significantly improves unlearning efficacy compared to existing baselines.

#machine-unlearning#privacy#multi-task-learning
📄
ArXiv AI71d ago

Hybrid KAN-MLP Architecture Boosts Human Activity Recognition Accuracy

Researchers propose a hybrid architecture combining Kolmogorov-Arnold Networks (KANs) and Multi-Layer Perceptrons (MLPs) to improve Human Activity Recognition (HAR). By leveraging KANs for input embedding and classification while retaining MLPs for feature mixing, the model achieves a 5.33% improvement in F1 scores over pure-MLP baselines.

#kan#har#neural-networks
📄
ArXiv AI71d ago

Evaluating LLM Utility for Personal Health Records

This study evaluates Gemini 3.0 Flash's ability to interpret complex Personal Health Records (PHRs) to answer patient queries. Results demonstrate that providing PHR context significantly improves the helpfulness, safety, and personalization of AI-generated responses.

#healthcare-ai#rag#llm-evaluation
📄
ArXiv AI71d ago

Developing Data Probes to Understand LLM Performance

This position paper proposes 'data probes'—synthetic sequences generated from random processes—to systematically analyze how specific data characteristics influence LLM behavior. This approach aims to replace empirical heuristics with a principled framework for understanding model training and inference.

#llm-training#synthetic-data#model-evaluation
Page 558 of 1667