"Six-Pack Abs" Aren't Just a Dream: AI Can Make It Come True

🧬 OpenAI GPT-4 High-Precision Modeling of Basic Protein Structures

🌏 Microsoft Phi-3.5 Series AI Models

🤖 Meta Launches New Web Crawler Bot

🚀 NVIDIA Unveils 8-Billion-Parameter AI Model

🎙️ Microsoft Azure AI Speech Service New Features

🛡️ World's First Automated Deepfake Detector

🌟 CarbonNovo: A New Era of AI-Driven Protein Design

🌐 China's Largest Intelligent Computing Center to Go Live This Month

🏢 Beijing Builds Global Robotics Industry Hub

🔍 Douyin Launches Fourth Search App

👥 Tencent Meeting Enhances Multilingual Translation

🤖️ Digital Huaxia "Xia Lan" Humanoid Robot

📈 LG Electronics Invests Tens of Millions in AI Startup

🚗 Chery Automobile Invests 100 Million Yuan to Found Kaiyang Technology

💰 Standard Bots Raises $63 Million

💼 "Dierboge" Completes Tens of Millions in Pre-A Round

💪 Gigabody AI Software: Easy "Muscle Gain"

OpenAI GPT-4 High-Precision Modeling of Basic Protein Structures

OpenAI's GPT-4 language model has demonstrated potential in high-precision simulation of amino acid and protein structures, according to research from Rutgers University.

GPT-4 can accurately predict the three-dimensional structures of 20 standard amino acids, including atomic composition, bond lengths, and angles. When simulating common protein structural elements such as α-helices, GPT-4 showed high consistency with experimentally determined structures. GPT-4 also successfully analyzed the binding between the antiviral drug Nirmatrelvir and the main protease of SARS-CoV-2, correctly identifying interacting amino acids and interatomic distances.

Microsoft Phi-3.5 Series AI Models

Microsoft's newly released Phi-3.5 series of AI models brings innovative Mixture of Experts (MoE) technology. All Phi-3.5 series models are now available on the Hugging Face platform under the MIT license.

Phi-3.5-MoE, the first Mixture of Experts model in the series, activates only 6.6 billion parameters in a 16 x 3.8B MoE architecture, demonstrating exceptional performance. It stands out in standard AI benchmarks, with performance approaching current leader GPT-4o-mini. Phi-3.5-vision has 4.2 billion parameters, enhanced understanding and reasoning capabilities for multi-frame images, significantly improving performance across multiple benchmarks. Phi-3.5-mini is a lightweight model with 3.8 billion parameters, supporting a 128K context window — far exceeding the 8K support of its main competitor, the Gemma-2 series.

Meta Launches New Web Crawler Bot

Meta's newly released web crawler, named Meta External Agent, was introduced last month. Functionally similar to OpenAI's GPTBot, it focuses on scraping text data for AI training, such as news articles and online discussions.

Mark Zuckerberg has stated that Meta's social platforms have accumulated massive AI training datasets, even surpassing Common Crawl. The launch of the new crawler may indicate that existing databases need updating and expansion to support the continued development of Llama models and Meta AI. According to Dark Visitors data, approximately 25% of popular websites globally have blocked GPTBot, while only 2% have blocked Meta's new crawler.

NVIDIA Unveils 8-Billion-Parameter AI Model

NVIDIA has achieved another breakthrough in AI with the release of the high-precision, high-computational-efficiency "Mistral-NeMo-Minitron 8B" AI model. "Mistral-NeMo-Minitron 8B" is a compact language AI model with 8 billion parameters. The model achieves GPU acceleration by running on workstations equipped with NVIDIA RTX graphics cards.

NVIDIA collaborated with Mistral AI to release the open-source "Mistral NeMo 12B" model last month; this 8B model represents further innovation building on that foundation. "Mistral-NeMo-Minitron 8B" performed excellently across nine popular language model benchmarks, including tests of language understanding, commonsense reasoning, mathematical reasoning, summarization, and the ability to generate factual answers.

Microsoft Azure AI Speech Service New Features

Microsoft Azure AI Speech Service's newly launched "Text to Speech Avatar" feature provides developers with a powerful tool for converting text to video. It is currently available in Southeast Asia, Northern Europe, Western Europe, Sweden Central, South Central US, and West US regions.

Developers can use this feature to transform text into videos that appear to be naturally spoken by humans, with output resolution of 1920×1080 at 25 frames per second, ensuring clarity and smoothness. Multiple preset character appearances are available, with virtual avatar voices powered by Azure AI text-to-speech technology for natural-sounding speech. It supports asynchronous or real-time synthesis of text-to-speech avatar videos, improving content production efficiency. Content creation tools are available in "Speech Studio," allowing users to create video content without writing code. Through Speech Studio's real-time chat avatar tool, real-time avatar conversation functionality can be enabled.

World's First Automated Deepfake Detector

The cybersecurity field has welcomed an innovative product: McAfee, in partnership with Lenovo, has launched "Deepfake Detector," effectively identifying and curbing the increasingly rampant AI scam activities. The tool's beta version has invited select Lenovo AI PC users in the United States, United Kingdom, and Australia to trial it, initially supporting English-language contexts.

"Deepfake Detector" has been trained on over 200,000 samples, possessing strong identification capabilities. The tool does not rely on cloud networks; instead, it utilizes the NPU in Lenovo AI PCs to perform Deepfake detection locally, effectively protecting user privacy. McAfee emphasizes that this local deployment approach can maintain user privacy and data security without sacrificing speed.

CarbonNovo: A New Era of AI-Driven Protein Design

The latest research from Haicang Zhang's team at the Institute of Computing Technology, Chinese Academy of Sciences — CarbonNovo — provides a novel solution for end-to-end de novo design of protein structures and sequences. CarbonNovo jointly designs protein backbone structures and sequences in an end-to-end manner through a unified energy-based model.

CarbonNovo is the first structure-sequence co-design model targeting all protein families, significantly exceeding baseline methods on metrics including foldability, diversity, and novelty. Network Recycling technology introduces protein language models into structural design for the first time, leveraging prior information from natural protein sequence data. CarbonNovo is applicable not only to protein monomer design but also has potential to extend to protein complex design and conditional design, such as peptides and antibodies.

China's Largest Intelligent Computing Center to Go Live This Month

China Mobile Intelligent Computing Center (Harbin) node's super-ten-thousand-card intelligent computing cluster will officially commence operations on August 30, placing Harbin among China's first-tier cities in AI research and development. The single-cluster computing scale is the largest domestically, with large-scale domestic network equipment deployment and leading integrated hierarchical storage, providing efficient and stable computing power support. Cluster failures can be located and bounded within minutes, with computing power reaching 6.6 EFLOPS.

According to the Guidelines for Innovative Development of Intelligent Computing Centers, intelligent computing center applications can significantly drive growth in AI core industries and related industries. During the 14th Five-Year Plan period, intelligent computing center investment can drive AI core industry growth by approximately 2.9-3.4x, and related industry growth by approximately 36-42x.

Beijing Builds Global Robotics Industry Hub

The 2024 World Robot Conference opened in Beijing's Yizhuang district, where Beijing Vice Mayor Jin Wei announced plans to build Beijing into a global robotics industry hub. Beijing will establish an embodied intelligent robotics innovation center, utilizing two open-source platforms — "Tiangong" and "Kaiwu" — as well as an embodied intelligent data entity. The innovation center will form a robotics shared processing center, promoting the construction and operation of precision component processing, concept verification, and pilot testing platforms.

Beijing will adhere to dual-wheel drive of product innovation and scenario demonstration, with coordinated industrial synergy and ecological optimization. By increasing the openness of application scenarios, implementing robotics new product engineering and application scenario demonstration engineering, and exploring "robotics+" application demonstration models, it will form a number of iconic scenarios, services, and business formats — such as robotics applications in healthcare, elderly care, landscaping, agriculture, and emergency response — driving industrial development.

Douyin Launches Fourth Search App

Douyin has intensified its exploration in search with the latest launch of the standalone app "Douyin Search," creating a search engine focused on its own content ecosystem. Douyin's development in search has been challenging, with previously launched Toutiao Search, Wukong Search, and Lightning Search failing to achieve expected success.

Unlike traditional search engines such as Baidu and Quark, "Douyin Search" primarily displays short videos and graphic content created by bloggers. Users can search by keywords to find short videos and graphic content under comprehensive channels, with segmented channels provided for precise search. "Douyin Search" avoids direct competition with traditional search engines, focusing instead on in-platform content search, leveraging its rich video and graphic resources. The addition of search functionality creates new commercial scenarios for e-commerce, local lifestyle services, and other businesses, helping to build a complete "view-search-purchase" ecosystem and improve conversion rates.

Tencent Meeting Enhances Multilingual Translation

Tencent Meeting has taken an important step in improving user experience and international services, introducing real-time translation in 17 languages to greatly facilitate communication and collaboration among users from different linguistic backgrounds.

Tencent Meeting's newly added real-time translation feature supports mutual translation among 17 languages including Chinese, English, Japanese, and Korean. The feature is available to Enterprise and Business edition users, and can be used in meeting subtitles, real-time transcription, and recorded playback. The translation functionality is powered by "Tencent Hunyuan Large Model" and "Tencent Translation Technology," ensuring accuracy and fluency.

Digital Huaxia "Xia Lan" Humanoid Robot

At the 2024 World Robot Conference (WRC), Digital Huaxia (Shenzhen) Technology Co., Ltd. showcased its innovative interactive humanoid robot "Xia Lan," which became a focal point with its highly realistic appearance, interchangeable faces, and natural voice interaction capabilities. The "Xia Lan" robot aims to break through traditional robot boundaries, creating "warm" interactive experiences and establishing emotional connections with humans.

Digital Huaxia has proposed the concept of "humanoid interactive robots," dedicated to creating intelligent companions capable of emotional resonance with humans. The "Xia Lan" robot's "intelligent brain" AI system can display rich expressions, with a highly realistic appearance close to that of real humans and a smooth control system. The robot is equipped with an "interchangeable face," enabling natural facial expression changes, as well as fluid "silky smooth movements."

LG Electronics Invests Tens of Millions in AI Startup

LG Electronics announced it will invest $10 million in the "Alpha Intelligence Fund," managed by SBVA, the early-stage venture investment arm of Japan's SoftBank. AIF is a newly established fund with a total scale of $130 million, focusing on AI and deep-tech startups. The collaboration between LG Electronics and SoftBank will combine both parties' resources and expertise to jointly identify and nurture promising AI startups. The investment is expected to accelerate the application and development of AI technology in consumer electronics, smart homes, enterprise solutions, and other fields.

Chery Automobile Invests 100 Million Yuan to Found Kaiyang Technology

Anhui Kaiyang Technology Co., Ltd. was established on August 15, with Tongyue Yin, Party Secretary and Chairman of Chery Holding Group, serving as legal representative. The registered capital is 100 million yuan, with registration in the Wuhu Economic and Technological Development Area, Anhui Province, adjacent to Chery Automobile's testing technology center.

Kaiyang Technology's business covers network and information security software development, intelligent robot R&D, artificial intelligence software development, cloud computing technology services, and more. The company also engages in frontier technology fields including quantum computing technology services, electronic specialty materials R&D, and emerging energy technology R&D. At the 2024 Lion Technology AI Agent Ecosystem Open Conference, Yin proposed that Chery's intelligentization will focus on two directions: Self Driving and Self Communication.

Standard Bots Raises $63 Million

"Standard Bots," a company focused on automation in manufacturing and logistics, has made significant progress in the AI robotic arm field and successfully secured substantial funding.

"Standard Bots" was founded in 2015 by Evan Beard and David Golden, dedicated to building easy-to-use yet powerful robotics platforms. The company's core competitiveness lies in its R&D capabilities and technical foundation, with a team composed of industry experts and top engineers. "Standard Bots'" AI robotic arm technology has broad application prospects in healthcare, manufacturing, and logistics, potentially driving industrial upgrading and transformation.

"Dierboge" Completes Tens of Millions in Pre-A Round

Dierboge, a company focused on intelligent robot process stations for the heavy industry sector, recently announced the completion of a tens-of-millions Pre-A funding round. Yuanhe Puhua led the round, with Toukong Donghai following, and Xinpei Capital serving as financial advisor. The funding will be used for technology product iteration and market expansion.

"Dierboge" was founded in 2022. Its core team comes from Tsinghua University, Xi'an Jiaotong University, and other institutions, with extensive experience in technology R&D and industrial implementation. They target steel, shipbuilding, aerospace, new energy manufacturing, and other fields, collaborating with multiple leading enterprises to provide standardized intelligent single-station delivery services including R&D, installation, and commissioning. Yu Chen, Managing Director at Yuanhe Puhua, is optimistic about Dierboge's team's intelligent algorithm capabilities and heavy industry experience. Chen Zhenhao, CEO of Toukong Donghai Shenzhen Bay Angel Phase III Fund, believes Dierboge possesses market scarcity, with product technology recognized by multiple leading customers.

Gigabody AI Software: Easy "Muscle Gain"

Gigabody, an AI software featuring one-click muscle enhancement, has rapidly gone viral. Since launch, it quickly climbed to second place on the free app charts, with over 150,000 daily downloads. Users simply upload an ordinary photo, select "one-click muscle enhancement," and within just 5 minutes obtain a robust, muscular physique. Despite Gigabody's stunning effects, some users report that the success rate is not entirely stable, sometimes requiring multiple attempts to achieve ideal results. The software offers one free transformation every 15 minutes, but additional filter effects require upgrading to the Pro version.

Sleep is humanity's true form of activity; being awake is merely collecting material for dreams.

— Shuji Terayama

Editorial Team Editor: Yuki

Design: Ivan

For business cooperation, please add WeChat: Rwkfbcianvd

References [1] Rutgers University research: https://www.nature.com/articles/s41598-024-69021-2

[2] "Mistral-NeMo-Minitron 8B" AI model: https://developer.nvidia.com/blog/mistral-nemo-minitron-8b-foundation-model-delivers-unparalleled-accuracy/

[3] CarbonNovo: https://openreview.net/pdf?id=FSxTEvuFa7

[4] Standard Bots: https://standardbots.com/

[5] Gigabody: https://top.aibase.com/tool/gigabody