Former Baidu executive Jing Kun founded Genspark and received 435 million in financing; CuspAI received 30 million US dollars in financing; Hinton served as a board advisor丨AI Intelligence Bureau
Latest update time：2024-06-20
Reads：
Intelligence Agency Live Report on CVPR2024
The 2024 CVPR conference was held in Seattle, USA, becoming the largest and most attended conference in the history of the conference, with a total of 12,000 participants. There were 35,691 registered authors in this conference, 11,532 papers were submitted, of which 2,719 papers were accepted, with an acceptance rate of 23.6%. Compared with last year, the number of papers increased by 20.6%, but the acceptance rate decreased slightly.
The conference awarded two best paper awards and two best student paper awards, including "Generative Image Dynamics" by the Google Research team and "Rich Human Feedback for Text-to-Image Generation" jointly published by multiple institutions. In addition, the conference also discussed hot topics such as visual basic models and image and video generation, as well as the latest research progress in areas such as machine forgetting, 3D vision and autonomous systems.
????https://mp.weixin.qq.com/s/45DYFWMXn-ce7ysJTrjp7g
AI Intelligence Bureau is looking for intelligence partners to collect exclusive valuable clues! If you can provide information about the latest AI achievements, industry insider information, and unique products, please add the operation WeChat account: AIyanxishe2 and note the industry position.
Financing Express
FINANCING NEWS
MainFunc receives $60 million in seed round financing:
MainFunc, founded by former Baidu executives (former Xiaodu Technology CEO Jing Kun and CTO Zhu Kaihua), has launched its first AI Agent search product Genspark. It has raised $60 million in an oversubscribed seed round of financing, led by BlueRun Ventures, valuing the currently unprofitable startup at $260 million.
CuspAI Raises $30 Million in Seed Round:
CuspAI, a developer of an AI-powered materials search engine, raised $30 million in seed funding led by Hoxton Ventures, with participation from Basis Set Ventures, Lightspeed Venture Partners, LocalGlobe, Northzone, Touring Capital, Giant Ventures, FJ Labs, Tiferes Ventures, and Zero Prime Ventures.
Point72 prepares new hedge fund focused on AI industry:
It is reported that Steve Cohen's Point72 Asset Management seeks to raise about $1 billion for a new stock picking hedge fund focused on AI. The fund will bet on AI hardware and semiconductor companies globally. This will be Point72's first new hedge fund in decades.
San Francisco AI Factory Inc received $20 million in financing:
San Francisco AI Factory aims to use AI to simplify coding tasks and provide automated AI systems - Droids, to help companies generate software features, review code, and resolve vulnerabilities and other engineering tasks. So far, Factory has raised a total of $20 million. In addition to Sequoia Capital, other investors include Lux Capital, Hugging Face, Databricks CEO, and Los Angeles music group The Chainsmokers.
Nvidia acquires software startup Shoreline:
Shoreline.io was founded by former Amazon Web Services executives. The valuation of Shoreline is approximately $100 million.
Constructor raises $25 million in Series B funding:
Constructor uses semantic search and artificial intelligence technology to provide accurate and personalized search results, and supports image, content and voice search products. This round of financing was led by Sapphire Ventures.
Aim Security Raises $18 Million in Series A Funding:
Aim Security focuses on the security of deploying and using generative AI tools in enterprise environments. This round of financing was led by Canaan Partners
Omi raises $14 million in seed funding:
Omi uses artificial intelligence to help brands create 3D visual assets, including still images and videos. This round of financing was led by Dawn Capital.
Finaloop Completes $35 Million Series A Funding:
Finaloop is an AI-driven e-commerce accounting platform. This round of financing was led by Lightspeed Venture Partners, with participation from Vesey Ventures, Commerce Ventures, and existing investors Accel and Aleph.
Aim Security Completes $18 Million Series A:
Aim Security is an enterprise AI security platform. This round of financing was led by Canaan Partners, and the company's seed round investor YL Ventures also participated in the investment.
Trustwise Raises $4M Seed Round:
Trustwise, a generative AI application performance and risk management startup, raised $4M in seed funding led by Hitachi Ventures, with participation from Firestreak Ventures and Grit Ventures.
Promaxo receives strategic investment
:
Promaxo is an American medical imaging service provider focusing on medical imaging, robotics and AI technologies. This investment was made by Zynext Ventures.
BioGeometry Completes Pre-A Round of Financing:
BioGeometry is a provider of open source machine learning platforms for macromolecular drug development. This round of financing was led by Jiangmen Venture Capital, followed by Zhipu AI and Shengjing Jiacheng, and old shareholder Gaorong Venture Capital continued to make additional investments.
HuanTian Wisdom completes Series B financing:
HuanTian Wisdom relies on remote sensing applications, cloud computing, big data, the Internet of Things, artificial intelligence and other information technologies to launch a system service architecture of "integration of sky, ground and space" and "space-cloud-network-terminal". The investor is CDH Capital.
Enveda Biosciences receives $55 million in financing:
Enveda uses its AI tools to identify and characterize the various molecules produced by organisms, thereby creating a new chemical biodiversity database. This round of financing was jointly participated by new investors Premji Invest, Lingotto Investment Fund, Microsoft, The Nature Conservancy and old shareholders Kinnevik, True Ventures, FPV, Level Ventures and Jazz Venture Partners.
Xianji Semiconductor Completes Nearly 100 Million Yuan Series B Financing:
Xianji Semiconductor is a domestic high-performance microcontroller manufacturer. This round of financing was led by Paradise Silicon Valley Capital, followed by Tianjin Yongtai Haihe, Hangzhou Yuanyan Equity Investment Fund and Sanwang Qitong. The financing will be used to accelerate the development of intelligent driving, robotics, edge AI chips and other fields.
(Welcome to add WeChat
AIyanxishe2
to learn more about AIGC and financing, and chat about new AI products with like-minded friends)
Industry News
INDUSTRY NEWS
Domestic Intelligence
Huawei Ascend AI computing power performance has surpassed NVIDIA A100, and nearly half of China's large models choose Ascend technology route:
Wang Tao, COO of Jiangsu Kunpeng·Ascend Ecological Innovation Center, said that the Ascend cluster is the only technology route in China that has completed the training of large models with hundreds of billions of parameters. The chip can reach up to 1.1 times the training efficiency of NVIDIA. "There is indeed a certain gap with NVIDIA A100 (referring to 0.8 times), but there is no obvious gap with NVIDIA A100 chip in large model training. Especially in the Wanka computing power cluster, including Kunpeng Cloud Brain and iFlytek, they have all been tested by the market."
The China Meteorological Administration released three AI meteorological model systems, named Fengqing, Fenglei, and Fengshun:
"Fengqing" is an artificial intelligence global short- and medium-term forecast system, and "Fenglei" is an artificial intelligence nowcast system. The two models were built by a research team formed by the China Meteorological Administration and Tsinghua University. "Fengshun" is an artificial intelligence global sub-seasonal-seasonal forecast system, which was built based on artificial intelligence methods by the China Meteorological Administration, Fudan University and Shanghai Institute of Science and Intelligence.
China Telecom and Zhiyuan released the world's first single dense trillion-parameter semantic model Tele-FLM-1T:
The model is based on technologies such as model growth and loss prediction. It uses only 9% of the computing power resources of the industry's general training solutions, 112 A800 servers, and completed the training of 3 models with a total of 2.3T tokens in 4 months. The TeleFLM series model has been fully open-sourced in a 52B version, with over 10,000 open-source model downloads and over 400,000 users. The Tele-FLM-1T version will also be open-sourced soon.
Baidu Xiling Digital Human Platform has been upgraded to support functions such as Vincent 3D Digital Human and voice cloning:
The new version of the platform can automatically generate realistic 3D digital people in a short time, and provides two cloning options: fast and high-quality to meet different needs. Fast cloning can be completed in half an hour, suitable for scenarios that pursue efficiency; high-quality cloning can restore real people 1:1, suitable for occasions with high requirements for real-life restoration. In addition, the Xiling platform has also launched a voice cloning function, which allows users to generate exclusive voices with only 30 seconds of recording.
Baidu Library's new product "Chengpian" supports the generation of 100,000-word long articles:
In terms of understanding extremely long pictures and texts, Chengpian can achieve lossless understanding of extremely long texts, support users to upload 100 files in multiple formats with a maximum size of 200MB each at one time, and support quick summary, Q&A and creation based on the uploaded content.
SenseTime disclosed that 50 papers were selected for CVPR 2024:
SenseTime Technology disclosed that 50 papers were selected for CVPR this year, of which 9 were accepted as oral or highlight. The papers involved cutting-edge fields such as autonomous driving and robotics.
International Intelligence
OpenAI and Color Health collaborate to create AI tools to assist in cancer screening and treatment:
OpenAI announced that it has partnered with genetic testing company Color Health to use the GPT-4o model to develop the AI tool Cancer Copilot to help doctors develop screening and treatment plans based on patient data, identify missing diagnostic results, and create tailored work plans, allowing healthcare providers to make evidence-based decisions about cancer screening and treatment.
TikTok launches new AI feature suite Symphony :
Symphony includes digital avatars, translation tools, AI assistants, and more. Brands can choose from a range of "stock avatars" based on real actors, or create custom avatars as virtual brand representatives. In addition, TikTok has launched the "Global Coverage Translation" feature. This is a new AI dubbing tool that can automatically transcribe, translate, and dub videos in more than 10 languages, helping brands expand their content globally.
Notion launches AI connector feature to improve workflow efficiency :
Users can extract knowledge directly from the company's Slack without leaving the current workflow, reducing the need to switch tools and windows. It has been released on the X platform and is designed to improve user productivity. Currently, Slack integration has begun to be gradually rolled out, and Google Drive and other undisclosed integration features are also being promoted.
Apple discontinues Vision Pro high-end machine:
Apple has suspended development of the next-generation Vision Pro and is focusing on releasing a cheaper model by the end of 2025. A low-priced Vision product called N109 may be launched, which is 1/3 the weight of Vision Pro and priced about the same as a high-end iPhone. It may retain a high-end display supplied by Vision Technology. The device has fewer cameras, a simpler headband, and smaller speakers.
Meta announced the reorganization of Reality Labs and the establishment of a new wearable device group:
After the reorganization, Reality Labs will be divided into two parts: one is the Metaverse: This department covers the Quest headset series, Horizon (Meta's social network) and related technologies. The other is wearable devices: This new department includes Meta's other hardware businesses, such as smart glasses in cooperation with Ray-Ban.
The Meta FAIR team released a number of models, research, and datasets:
The Meta FAIR team released a number of models, research and data sets, including Meta Chameleon: multimodal model, 7B/34B; Multi-Token Prediction: multi-word prediction model; JASCO: text-generated music model; AudioSeal: AI speech detection; PRISM: AI feedback data set; "DIG In": human geography difference assessment method.
Universal Music and SoundLabs launch AI plug-in MicDrop:
MicDrop is an AI vocal plug-in that uses the artist's own voice data for training to create a high-fidelity vocal model that retains the artist's ownership and is used for exclusive creations and is not available to the public. It will be launched this summer and is compatible with all major DAWs. Universal Music says it can achieve a variety of sound conversions.
Team training AI, MLX project debuts:
The MLX project uses MPI distributed computing and connects the host computer and multiple Mac devices through Thunderbolt 4 cables, which can achieve efficient parallel computing and is suitable for scenarios such as training AI in a home environment. Apple has previously explored and developed a similar XGrid project, connecting multiple Mac devices in series to achieve parallel computing, but it is mainly aimed at enterprises and government agencies, and is not friendly to consumers and amateurs.
Hinton, the “Godfather of AI”, serves as an advisor to the CuspAI board of directors:
Hinton spoke highly of the startup, saying he was impressed by the company and its mission. "They are using AI to speed up the process of designing new materials to address one of humanity's most pressing challenges - climate change." CuspAI, founded by Cambridge University, plans to use the power of search engines to identify the properties required for new building materials on demand.
Models such as ChatGPT are being trained madly, and the AI industry may face a "data shortage" in 2026:
The Epochai research report points out that currently there are about 300 trillion tokens of high-quality text training data sets available to the public, but as the appetite of large models grows, these data may soon be exhausted. For example, Meta's Llama3 model was overtrained by an astonishing 100 times on the 8B version.
Epochai proposed four methods to obtain new training data: synthetic data, multimodal and cross-domain data learning, the use of private data, and real-time interactive learning with the real world. The aim is to avoid the "data shortage" in the AI community and provide data support for the continued development of AI models.
The best papers of ACM's top conference SIGGRAPH 2024 were announced, with NVIDIA and CMU each accounting for 40%:
ACM SIGGRAPH selected 5 best papers and 12 honorable mentions, and continued last year's tradition by awarding the Test of Time Award to 4 papers published in 2012 and 2013. Domestic institutions such as ShanghaiTech University, Huazhong University of Science and Technology, and the Chinese University of Hong Kong were on the list.
More international information
BCG report says generative AI is shaking up the job market:
The report predicts that generative AI will have an economic impact of at least $2.2 trillion to $3.7 trillion on the global economy in the next decade, posing a threat to certain highly repetitive and low-creativity jobs, but will also create new jobs and drive talent to reshape and learn their skills.
Samsung Electronics to launch AI-equipped home appliances next year:
Samsung Electronics is developing integrated home appliances with large language models, aiming to release them in 2025.
AI Products
AI PRODUCTS
Product Hunt hot list, free AI agent search engine Genspark :
Genspark is a free AI agent search engine that uses professional AI agents to research and generate so-called Sparkpages for user queries. These pages integrate reliable information, provide more valuable results, and save time for users. Founder Jing Kun emphasized that Genspark is different from traditional search engines, and is more like a group of useful AI partners who can quickly find the answers users need. Genspark aims to eliminate advertisements, misleading content, and biased results, and provide clean, high-quality information, allowing users to access the information they need from one place and save time.
????https://www.genspark.ai/
GitHub Trending hot list, open source enhanced ChatGPT clone LibreChat:
LibreChat is an open source enhanced ChatGPT clone project that supports multiple AI models and APIs, including OpenAI, Azure, Groq, etc. It has features such as AI model switching, message search, and multi-user security system, and is actively developing more features. LibreChat, maintained by danny-avila, has 12.9k stars and 2.3k forks on GitHub.
????https://www.librechat.ai/
????https://github.com/danny-avila/LibreChat?tab=readme-ov-file
3. Hedra Labs launches Character-1 research preview
Hedra Labs has released a research preview of Character-1, a basic model that can generate expressive speaking, singing and rapping characters. The model can be used on desktop and mobile devices. The preview version provides unlimited video duration, but the open preview version is limited to 30 seconds of video. If the H100 supply is sufficient, the model can generate 90 seconds of video every 60 seconds. The model has the characteristics of strong expressiveness of the generated characters. Its vision is to inspire the next generation of human storytelling ability by building a basic model and incorporating it into products. It also announced the upcoming "Worlds" feature that allows users to build virtual worlds.
????https://www.hedra.com/
4. GenType creates a custom alphabet
GenType is an online tool that uses the Imagen 2 API to provide users with the ability to create custom alphabets. Users can customize the style of the alphabet by describing it, such as using elements such as constellation maps, futuristic sci-fi spaceships, and silver pipes. GenType reminds users to respect the rights of others when creating, and encourages users to share feedback to help improve AI.
????https://labs.google/gentype
Insights from the Big Bull
DEEPING SAYING
The most powerful GenAI/LLM learning resource index is released!
Will Brown published the GenAI Handbook, which is regarded as the most cutting-edge open source textbook in the field of GenAI. It brings together the development and systematic knowledge guide of the GenAI/LLM field in the 18 months since the release of ChatGPT. It is divided into 9 parts, with references to top blogs, papers, Youtube videos and online courses, providing readers with a clear understanding of the development of GenAI.
????https://genai-handbook.github.io/
Popular Activities
On June 28, the Shenzhen stop of "Attent!on" will focus on the integration of software and hardware, with the theme of "AI+cross-border+hardware=?", to explore the opportunities and challenges of AI hardware.
Yunqi Capital and Leifeng.com will work together with Zhixing Institute, an innovative enterprise empowerment organization founded by Gan Jie, an early incubation investor of DJI and professor of finance at the Cheung Kong Graduate School of Business, to engage in in-depth exchanges with senior people from well-known companies such as Huawei, Tencent, iFlytek, Kickstarter, Yuansheng Intelligence, Huohuotu, Time Space Pot, and Honeycomb Technology.
Entrepreneurs and product managers are welcome to sign up.
???? https://mp.weixin.qq.com/s/D9YIyKBz0UUdjP3iNqXefA
Stay tuned for the latest updates tomorrow
!
AI Intelligence Bureau is looking for intelligence partners to collect exclusive valuable clues! If you can provide information about the latest AI achievements, industry insider information, and unique products, please add the operation WeChat account:
AIyanxishe2
and note the industry position.
//
Recent Hot Articles
京公网安备 11010802033920号