27
When Structured Sparse Autoencoders Learn Consistent Concepts Across Modalities
Researchers propose Structured Sparse Autoencoders to help vision-language models learn consistent, non-fragmented concepts across both visual and textual modalities.
27
Fine-tune NVIDIA Nemotron 3 models with Amazon SageMaker AI serverless model customization
Amazon Web Services has introduced serverless customization capabilities on SageMaker AI to simplify the fine-tuning of NVIDIA's Nemotron 3 models.
26
2026年度「中国股权投资行业投资机构」系列名册正式发布
A newly released report on China's venture capital sector highlights a major shift toward hard tech, with embodied AI alone attracting over 65.6 billion yuan in funding.
26
Hugging Face’s CEO on why companies are done renting their AI
Hugging Face's chief executive argues that businesses are increasingly shifting away from proprietary AI subscriptions in favor of open-source alternatives.
26
Two Axes of LLM Abstention: Answer Correctness and Question Answerability
The study reveals that large language models treat answer correctness and question answerability as two distinct dimensions, suggesting that a single confidence score is insufficient for proper model abstention.
26
PLURAL: A Global Dataset for Value Alignment
Researchers have released PLURAL, a large-scale dataset based on global survey data designed to help align language models with diverse international value systems.
26
韩国股指年内大涨八成估值却创低位，机构称存储板块仍具备配置价值
Despite a surge in South Korean stock indices driven by soaring AI-related profits at Samsung and SK Hynix, historically low valuations present a potential entry point for investors.
26
DJI launches first VTOL fixed-wing cargo drone EV50, says it set Everest altitude record
DJI has introduced the EV50, its first vertical takeoff and landing fixed-wing cargo drone, which reportedly set an altitude record during a scientific expedition on Mount Everest.
26
Equinix to develop 10th and 11th data centers in Amsterdam
Equinix is planning two new data centers in Amsterdam, though local restrictions will limit construction to a quarter of the permitted development capacity.
26
SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric Scheduling
SMetric is a scheduling framework designed specifically for agentic workloads, optimizing serving efficiency by prioritizing complete agent responses and leveraging high KV-cache reuse.
26
Microsoft’s patch Tuesdays are about to get bigger
Microsoft is integrating artificial intelligence into its Windows 11 update pipeline to identify and patch security vulnerabilities more rapidly.
26
Huawei Mate 90 Series Confirmed: Debuts HarmonyOS 7 and Tao's Law Kirin Chip, Taking on iPhone 18 Pro
Huawei has confirmed its upcoming Mate 90 smartphone series will feature HarmonyOS 7 and a new Kirin chip utilizing advanced transistor density technology.
25
Real-time dental image verification with Amazon SageMaker AI at Henry Schein One
Henry Schein One has deployed an AI-powered quality verification system on Amazon SageMaker to analyze dental X-rays in real time across thousands of clinics.
25
Chinese AI labs pursue custom chips to lower costs but heavy upfront investment a risk
Chinese artificial intelligence laboratories are increasingly developing proprietary chips to improve hardware-software integration and reduce long-term operational costs despite high upfront financial risks.
25
Deploying quantized models on Amazon SageMaker AI with Unsloth
AWS has outlined four deployment patterns for running quantized machine learning models optimized by Unsloth on its SageMaker and cloud container infrastructure.
25
Tiny robot boats build floating structures
MIT researchers have developed FloatForm, a swarm of small aquatic robots capable of self-assembling into reconfigurable floating structures.
25
WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search
WebSwarm is a recursive multi-agent framework designed to improve the depth and coverage of web search tasks by dynamically orchestrating specialized sub-agents.
25
DeepSeek begins in-house AI chip development to cut reliance on NVIDIA, sources say
Chinese artificial intelligence startup DeepSeek has initiated an in-house chip development project focused on optimizing inference workloads and reducing reliance on Nvidia hardware.
25
Disaggregated prefill and decode for LLM inference on SageMaker HyperPod
AWS has demonstrated how to implement disaggregated prefill and decode using vLLM on SageMaker HyperPod to optimize large language model inference.
25
Cognitive-structured Multimodal Agent for Multimodal Understanding, Generation, and Editing
To prevent context window overload in long-horizon multimodal dialogues, this paper introduces an agent architecture that stores visual information in an external memory structure.
25
Prismata: Confining Cross-Site Prompt Injection in Web Agents
The Prismata framework is designed to secure autonomous web agents against cross-site prompt injection attacks by isolating untrusted web content from core agent instructions.
25
When Implausible Tokens Get Reinforced: Tail-Aware Credit Calibration for LLM Reinforcement Learning
The paper identifies a credit assignment failure in critic-free reinforcement learning for language models and proposes a tail-aware calibration method to address it.
25
Rethinking LLM-as-a-Judge: Representation-as-a-Judge with Small Language Models via Semantic Capacity Asymmetry
Researchers propose a Representation-as-a-Judge method that leverages the internal representations of small language models to evaluate outputs efficiently without relying on text generation.
25
Can AI answer the $3 trillion question?
This article explores the ongoing debate surrounding the return on investment for massive capital expenditures in the artificial intelligence sector.