news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Today's Financing Express

Artificial intelligence chip startup Groq receives $640 million in funding to challenge Nvidia

Groq, a startup developing chips to run generative AI models faster than traditional processors, said Monday it has raised $640 million in a new round of funding led by Blackrock. Neuberger Berman, Type One Ventures, Cisco, KDDI and Samsung Catalyst Fund also participated.

The funding brings Groq’s total raised to over $1 billion and values ​​the company at $2.8 billion, a major win for Groq, which had reportedly initially been looking to raise $300 million at a slightly lower valuation of $2.5 billion.

Yann LeCun, Meta’s chief AI scientist, will serve as a technical advisor to Groq, and Stuart Pann, former head of Intel’s foundry business and former chief information officer of HP, will join the startup as chief operating officer.

UDI Robotics was transformed into a joint-stock company and its capital was increased to RMB 350 million

UDI Robotics (Wuxi) Co., Ltd. has undergone industrial and commercial changes, with the market entity type changed from a limited liability company (invested by Hong Kong, Macao and Taiwan, non-sole proprietorship) to a joint stock company (invested by Hong Kong, Macao and Taiwan, unlisted), and the name changed to UDI Robotics (Wuxi) Co., Ltd. At the same time, the registered capital increased from approximately RMB 17.079 million to RMB 350 million. The official website shows that UDI Technology focuses on the application research and development and commercialization of the core technology of delivery robots.

Robotics startup DELIVERS.AI is valued at $36 million after new funding round

DELIVERS.AI’s autonomous mobility platform uses advanced, AI-driven, low-emission on-road delivery robots and vehicles to make last-mile logistics affordable and sustainable.

DELIVERS.AI has raised an undisclosed amount of funding at a valuation of $36 million. The Warwick, UK-based company received funding from Japan Post Capital, Turkey Development Fund, Bulgaria’s Boost Capital, and Istanbul Technical University. Previous investors include Driventure, Arz Portföy, StartupFON, Plug and Play Ventures, Inveo Ventures, StartersHUB, and Kalyon Ventures.

AI-driven jewelry company Stepin receives 10 million yuan in angel round financing

Stepin is a cross-border jewelry brand that focuses on AI jewelry cross-border e-commerce. With the help of AI technology and China's supply chain advantages, it realizes the "small order, fast return" model of rapid new products and high turnover. Stepin has completed a 10 million yuan angel round of financing, led by Xinyue Capital and Jiujiu Capital, with other industry institutions and angel investors participating in the follow-up investment, and Inspur Capital serves as the exclusive financial advisor. At present, the application of AI within Stepin is mainly to generalize creativity and generate relatively basic 3D models for designers to fine-tune. (36Kr)

(Welcome to add WeChatAIyanxishe2, learn more about AIGC and financing, and chat with like-minded friends about new AI products)

Today's big factory news

ResNet author Zhang Xiangyu reportedly joins Step Star

According to QuantumBit, Zhang Xiangyu, a post-90s AI expert, has joined Step Star. He is one of the four authors of ResNet, Sun Jian's first deep learning doctoral student, and the winner of the Future Science Award. Zhang Xiangyu received a bachelor's degree in software engineering from Xi'an Jiaotong University, and worked with Sun Jian, He Kaiming, and Ren Shaoqing at Microsoft Research Asia to complete ResNet. The paper won the CVPR Best Paper Award in 2016 and won the "Mathematics and Computer Science Award" of the Future Science Award in 2023.

In addition to Zhang Xiangyu, Jieyuexingchen also recruited two other Wanyin experts, Tencent's Yu Gang and MSRA's Duan Nan. Yu Gang is an undergraduate at Shandong University, a master's degree at Shanghai Jiaotong University, and a doctorate at Nanyang Technological University. He has interned at Microsoft Research and completed a number of research results while working at Megvii. Duan Nan is a senior principal researcher at MSRA, mainly engaged in research such as natural language processing, and has worked at MSRA for 17 years and 9 months. His research results have been applied to a number of Microsoft AI products.

OpenAI admits it is developing ChatGPT text watermarks, but faces challenges

OpenAI has developed a tool that can recognize ChatGPT-generated text with high accuracy, but it has not yet been released. In response, OpenAI admitted that it is researching text watermarking technology, but said that this technology still has many challenges.

OpenAI envisions weaving an invisible "digital fingerprint" - a text watermark - between the lines by subtly adjusting the vocabulary choices in the text generated by ChatGPT. The cleverness of this design is that in the future, it will be easy to identify and verify the original source of the text with the help of specific tools, opening up new paths for copyright protection and content traceability. And text watermarks are only one part of OpenAI's diversified solution matrix. They are also studying classifier technology and metadata strategies in parallel, aiming to build a comprehensive, multi-level text identity authentication system to ensure that the source of information is clear and traceable.

Figure previews the second generation of humanoid robots, with more human touch and more powerful hardware

Figure released a trailer for Figure 02 and announced that it will officially release the product on August 7, Beijing time. Compared with the video demonstration of Figure 01 equipped with Open AI GPT4, the focus of this demonstration is on hardware, and it is expected that the hardware capabilities will be greatly improved. Founder and CEO Brett Adcock confidently said: Figure 02 is the best humanoid robot on earth.

Google Gemini API price cuts to half the price of GPT-4o mini

The input cost of the Gemini 1.5 Flash model has been halved, down about 85%, and the output cost is close behind, slashed by about 80%. This means that the cost of using the Gemini API is now nearly 50% lower than its main competitor, GPT-4o mini. The new pricing of Gemini Flash costs only $0.075 per million tokens for input and only $0.3 for output. Gemini 1.5 Flash and Gemini 1.5 Pro now support more than 100 languages, and Google has also introduced innovative technologies such as context caching and batch APIs.

Baichuan Intelligence and Renmin University of China established a "Large Model Joint Laboratory"

Renmin University of China and Baichuan Intelligence jointly established the "Big Model Joint Laboratory" to promote the innovation and development of big model technology. The establishment of the joint laboratory marks that the two parties will carry out in-depth cooperation in cutting-edge technology fields such as big model pre-training, alignment, retrieval enhancement, intelligent agents, and multimodality. Renmin University of China will use its talent and technical advantages in big model research, combined with Baichuan Intelligence's strength in engineering and product development, to jointly promote the research and application of related technologies.

Alibaba launches Tora, the "trajectory-controllable version of Sora" to make video generation more in line with the laws of physics

Tora is the first trajectory-oriented DiT architecture that simultaneously integrates text, vision, and trajectory conditions to generate videos. Tora's design fits seamlessly with DiT's scalability, allowing precise control of video content with different durations, aspect ratios, and resolutions. Extensive experiments demonstrate that Tora excels in achieving high motion fidelity while also carefully simulating motion in the physical world.

Meta is reportedly negotiating AI voice projects with Hollywood stars and will offer millions of dollars for licensing

According to media reports citing sources, negotiations between Meta and some actors' representatives have been interrupted and restarted several times because the two sides could not agree on the terms of use of the actors' voices. Meta is accelerating the negotiations to have enough time to develop AI tools, hoping to release them at the Connect conference in September. It is not clear how Meta will use these voices, most likely as a digital assistant. For example, users can chat with a chatbot with Awkwafina's voice.

todayProduct News

Product HuntHot List, Avatar Architect

Avatar Architect is a system that combines artificial intelligence and Notion. It aims to help entrepreneurs, marketers and product developers improve marketing strategies and sales performance by gaining a deep understanding of target customer groups, while providing a series of tools and guides to build and manage customer portraits.

Avatar Architect's strengths lie in its AI-driven efficiency, deep insights into target markets, and detailed customer information to guide product development. The system is suitable for independent entrepreneurs, marketers, and product developers, especially those who want to gain a deeper understanding of their customers and improve their marketing strategies. Users can customize the system's functionality according to their business needs, and customer data should be updated regularly for best results.

https://gcproductivity.gumroad.com/l/avatararchitect/ProductHunt?ref=producthunt

Developer Recommendation

1. Simple tips to easily create small program code LlamaCoder

LlamaCoder is a platform based on the Llama3.1405B model. Through powerful automation capabilities, it allows developers to quickly generate complete React applications and components by simply providing simple instructions. The platform uses a modern technology stack, including popular technologies such as Next.js and Tailwind, and provides an interface that is both beautiful and easy to use. LlamaCoder's functional design takes into account comprehensive considerations, including code sandboxes, Helicone integration, and the use of Plausible tools to improve development efficiency and product optimization. The entry threshold is low, and users only need to clone the code base and set the API key to start the project through the npm command. It is now open source.

https://llamacoder.together.ai/

2. Supermemory Personal Knowledge Base Project

The project allows users to save online information such as web pages, tweets, and notes, and use its built-in artificial intelligence functions to search and ask questions efficiently. It organizes information in the form of a two-dimensional canvas to help users better understand and associate knowledge points. It provides AI-assisted writing functions based on saved data. It supports integration with platforms such as Telegram and Twitter.

github https://github.com/supermemoryai/supermemory

Website: https://supermemory.ai/onboarding

Special attention

Jim Fan: Amplification of robot data is a key issue in solving the development of robotics technology

Jim Fan, senior research scientist at NVIDIA, head of embodied intelligence and head of Embodied AI (GEAR Lab), released the latest progress of Project GR00T, proposing a systematic approach to amplify robotics datasets, using human demonstrations on real robots combined with simulation technology to increase the amount of data by 1,000 times or more to solve the data bottleneck problem in robotics technology.

Apple Vision Pro technology is used to enable human operators to control humanoid robots from the first person. Vision Pro can interpret human hand gestures in real time and map them to robot hands, making the operator feel immersed in another body. Although remote operation is slow, a small amount of high-quality data can be collected.

Using RoboCasa, an open-source generative simulation framework, we can generate a large amount of diverse demo data by extending a single real-world demo data into a wide variety of environments by changing the visual appearance and layout of the environment. This allows the data of a physical kitchen to be extended to an unlimited number of simulated kitchen scenarios.

Apply MimicGen technology to generate a large number of new motion trajectories based on the original human demonstration data and filter out failed attempts, resulting in a larger and richer dataset.

With this approach, starting from a human trajectory, RoboCasa can generate data for N different visuals, which MimicGen further enhances to data for NxM different actions. This approach solves the problem of expensive human data collection at the expense of computational power through GPU-accelerated simulation, breaking the traditional data collection barrier limited to the atomic world.

https://x.com/DrJimFan/status/1818302152982343983

Please look forward to the latest updates tomorrow！

AI Intelligence Agency 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:AIyanxishe2Note the industry position.

Google acquires CharacterAI for over $2.5 billion; Nvidia's AI chips face major design flaws, billions of dollars in orders will be affected | AI Intelligence Agency

Estun Robotics received an additional investment of 450 million yuan; AI expert Zhou Zhihua was appointed as vice president of Nanjing University; papers can also be posted with comments! Stanford's online paper platform is popular | AI Intelligence Bureau

Stardust Intelligence received tens of millions of dollars in financing, focusing on the commercialization of AI robots; OpenAI partially opened GPT-4o voice, and expanded to all paying users this fall丨AI Intelligence Bureau