AI Trending Daily Digest

China's humanoid robots account for over 80% of the global market

China's robotics sector achieved a significant export surge in the first quarter, with robot shipments valued at 11.32 billion yuan reaching 148 countries. The nation holds a commanding global lead in humanoid robot production, accounting for over 84% of the world's supply from a base of 140 companies. This industrial foundation is projected to fuel the domestic embodied intelligence market to 400 billion yuan by 2030 and potentially one trillion yuan by 2035.

Score: 59

Two AI Series Break into Cannes: Jellyfish Intelligence Aims to Be the Breakthrough Player in the Industrialization of AI Content | Project Report

Two fully AI-produced vertical dramas from Waterdrop Intelligent’s studios were selected for the Cannes Film Festival’s Fantastic Pavilion, marking a

Score: 57

How to Build Programmable AI Agents on Snowflake? In-depth Analysis of Cortex Code Agent SDK | Technical Practice

I see the prompt and format, but no article content was actually provided. The message shows "点击查看原文>" which appears to be a link or placeholder rather than the text itself.

Score: 56

CSC Financial: The core investment logic of the overseas AI industry presents two main themes

Global AI investment is splitting into two clear tracks: a hardware-infrastructure shift and a model-cloud ecosystem realignment. On the hardware side

Score: 56

GlobalWafers plans to raise prices in the second half of the year.

GlobalWafers chairperson Hsu Hsiu-lan said market conditions this year are **significantly better** than last year, supported not only by strong AI-re

Score: 55

The beta version of Grok Build under xAI is now available to SuperGrok and X Premium+ users.

xAI has expanded Grok Build beta access to all SuperGrok and X Premium+ users, positioning it as a broader creation and automation tool rather than a

Score: 55

A Conversation with Yunjilink CEO Zhang Junbin: Understanding Human Nature, Bidding Farewell to Naivety

At 36, Yunjing CEO Zhang Junbin realized his gentle, product-obsessed nature had hindered his leadership, allowing internal decay through departmental friction, talent loss, and inefficient marketing. Forcing himself to become a decisive "ruthless" leader, he took direct control, exposed flawed metrics and siloed thinking, and fundamentally shifted his view from "Product + Marketing" to "Product x Marketing." This painful evolution involved confronting human complexity, learning to make tough pe

Shares traded in Hong Kong of DeepGlint surged over 140% in pre-market trading.

Shenyuan Intelligent's stock price surged over 140% on the Hong Kong grey market, trading at 135.1 HKD, ahead of its official listing on the Hong Kong Stock Exchange scheduled for May 27th.

Snowflake AI Guide: Boost Development Efficiency and Reduce Token Costs | Technical Practice

The city of Chengdu has implemented a ban on 22 dog breeds deemed dangerous, including Tibetan Mastiffs and Pit Bulls, effective December 1, 2023, following a series of incidents where dogs attacked children. Violations will result in fines and potential confiscation of the animals. The regulation aims to enhance public safety but has sparked debate over breed-specific legislation versus responsible pet ownership.

Galaxy Securities: Tokens Become the Value Anchor of the Intelligent Era; It Is Recommended to Position Around Core AI Industry Beneficiary Segments

Tokens have become the core unit of value in the intelligent era, serving as the standard settlement measure across the AI industry while surging glob

AI Products

Ferrari Luce

LoveFrom, the design studio co-founded by Jony Ive, has revealed its first electric vehicle design for Ferrari, marking a significant entry for the iconic Italian automaker into the electric era. The design prioritizes a seamless, sculptural form language that blends Ferrari's legendary performance ethos with the inherent packaging advantages of an electric powertrain, aiming to create a new emotional and aesthetic paradigm for the supercar.

Score: 52

Parameter Efficient Multi-Class Intelligent Scheduling for Multimodal Online Distributed Industrial Anomaly Detection

The MODIAD framework addresses the gap in industrial anomaly detection for distributed, continuously generated multimodal data. It proposes a complete workflow involving a Multi-class Intelligent Scheduling (MIS) algorithm to coordinate model updates and a Resource Efficient Class-Wise Low Rank Adaptation (REC-LoRA) strategy to minimize computational overhead. This enables efficient, collaborative detection on edge devices, demonstrating superior performance on standard datasets under realistic

Score: 52

CAFD: Concept-Aware DNN Fault Detection using VLMs

Concept-Aware Fault Detection (CAFD) introduces an efficient learning-based method for identifying Deep Neural Network faults by uniquely combining model outputs, distance-based features, and a novel concept-based metric called the Concept Failure Ratio (CFR). CFR uses Vision-Language Models to extract and quantify the link between semantic concepts in images and DNN failures, providing complementary information that boosts detection accuracy. Evaluations show CAFD consistently outperforms five

Score: 52

AI Products

MiniCPM5-1B

This article announces a significant breakthrough in developing compact, open-source models optimized for edge devices, achieving state-of-the-art performance in their class. The core innovation likely involves novel architectural or training techniques that dramatically improve efficiency without sacrificing accuracy, enabling advanced AI capabilities to run locally on resource-constrained hardware like smartphones or IoT devices, thus enhancing privacy and reducing latency.

Score: 50

A Large-Scale Dataset and Benchmark: Do Protein-Ligand Models Learn Binding Sites or Just Binding Likelihood?

The paper introduces InteractBind, a large-scale benchmark for evaluating protein-ligand modeling that moves beyond traditional binary binding prediction and affinity regression. Its core contribution is a fine-grained assessment of **binding-site localization** using detailed interaction maps for six non-covalent interaction types, revealing that existing models perform poorly at this task despite being accurate at binding prediction. This establishes a new paradigm for developing more interpre

Score: 50