Baichuan Intelligence completes 5 billion yuan Series A financing, with state-owned capital from Beijing, Shanghai and Shenzhen entering the market; OpenAI challenges Google and releases AI search SearchGPT丨AI Intelligence Bureau
Latest update time：2024-07-26
Reads：
Today's Financing Express
Baichuan Intelligent completed RMB 5 billion in Series A financing, with state-owned capital from Beijing, Shanghai and Shenzhen entering the market
Baichuan Intelligence is one of the first domestic AI large-scale model companies established in April 2023, founded by former Sogou CEO Wang Xiaochuan. Its core team consists of AI talents from Sogou, Google, Tencent, Baidu, Huawei, Microsoft, ByteDance and other technology companies.
Baichuan Intelligence has completed its A round of financing, with a total financing amount of 5 billion yuan, and will start its B round of financing with a valuation of 20 billion yuan. The investors in the A round include Alibaba, Xiaomi, Tencent, Asia Investment Capital, CICC and other leading companies and market-oriented investment institutions, as well as state-owned industrial investment funds such as Beijing Artificial Intelligence Industry Investment Fund, Shanghai Artificial Intelligence Industry Investment Fund, and Shenzhen Capital Group. (36Kr, Jiemian News)
Harvey announces completion of $100 million Series C financing from GV, OpenAI, and Sequoia Capital USA! Latest valuation: $1.5 billion
On July 23, Harvey officially announced the completion of its Series C financing led by Google Ventures (GV), with participation from OpenAI, Kleiner Perkins, Sequoia Capital, Elad Gil and SV Angel. It is reported that the total amount of this round of financing reached US$100 million, and the company's valuation was US$1.5 billion. This money will enable Harvey to expand and improve its AI technology globally. Harvey plans to use this new capital to invest in engineering, data, and domain expertise, which are essential for building AI-native systems that facilitate the most complex knowledge work. (Newin)
Whale Robotics completes its seventh round of financing, with a financing amount exceeding 100 million
Whale Robot is a global high-tech enterprise focusing on the field of AI Robot + Edu, determined to provide AI Robot products for 1.5 billion young people around the world. It integrates the design, development, production and sales of robot software and hardware.
Whale Robotics announced the completion of its seventh round of financing. The Shanghai Sports Industry Fund, jointly established by multiple state-owned capital platforms including Shanghai Jiushi Group, Lujiazui Group, Shanghai Media Capital, and Five Star Sports Media, participated in this round of financing, with a financing amount of more than 100 million yuan. (36Kr)
AI threat detection startup Lakera AI
raises $20 million in Series A funding
Generative AI threat detection startup Lakera AI AG, known for creating one of the world’s largest “AI red teams” to test generative AI models, today said it has closed $20 million in early-stage funding. Lakera also has a second product called MLTest, which is designed for developers who need to test the performance of computer vision models before deploying them in production.
The Series A round was led by European venture capital firm Atomico, with participation from Citi Ventures, Dropbox Ventures, and Redalpine, bringing the startup’s total funding to $30 million
Radiology AI startup Deepc raises $13 million in extended Series A funding
Founded in 2019, Deepc, through deepcOS, provides clinicians with access to regulatory-approved AI solutions for over 60 clinical indications, enabling healthcare providers to benefit from the rapid advancements in AI in radiology without any of the technical, operational or commercial complexities.
Deepc has completed a Series A funding round with an extension of $13 million, bringing its total funding to date to $30 million. Sofinnova Partners and Bertelsmann Investments co-led the round, with participation from new investors KHP Ventures and SwissHealth Ventures, and existing investor Winning Mindset Ventures.
AI-driven enterprise learning company Uplimit completes $11 million Series A funding
Uplimit, a leader in AI-driven enterprise learning, announced that it has closed $11 million in Series A funding. Salesforce Ventures led the round alongside existing investors GSV Ventures, Greylock Partners, and Cowboy Ventures, as well as new investors Translink Capital, Workday Ventures, and Conviction.
Medical robot developer Clarapath completes Series B financing
Founded in 2014, Clarapath is a medical robot developer dedicated to providing integrated technologies to automate and enhance the processing of human and animal tissues for pathological examination. Clarapath announced the completion of its Series B financing, with investors including CU Healthcare Innovation Fund, Mayo Clinic, Northwell Ventures and Ochsner Ventures.
AI trust startup Vijil receives $6 million in seed funding
AI safety startup Vijil AI Inc. has raised $6 million in seed funding as it launches its first cloud-based tool to help companies build and deploy more reliable generative AI agents. Vijil only requires customers to provide a small sample of data about how their model is used, and then creates a comprehensive test suite to measure how it performs in nearly any type of scenario in that context and scores the model. It works with a variety of generative AI systems, including open source large language models, closed AI APIs, retrieval-enhanced generation applications, and AI agents.
The round was co-led by Mayfield LLC’s AIStart Seed Fund and Google LLC’s AI-focused seed fund, Gradient Ventures.
AI-driven cloud risk solution startup ZEST Security raises $5 million
ZEST provides a platform that correlates and pinpoints the root causes of cloud risks and develops resolution paths to eliminate cloud vulnerabilities and misconfigurations that attackers can exploit. Funding comes from Hanaco Venture Capital Ltd., Silvertech Ventures LP, and some unnamed angel investors.
AI no-code workflow platform Gumloop receives $3.1 million in funding
Gumloop was founded more than a year ago as a side project for a niche user group in the Discord community, helping them automate their work using AI. It has now grown into an automation platform that allows anyone to fully automate their work with drag and drop. Users can create powerful AI-driven workflows that integrate directly with their data without programming experience.
AI automation platform Gumloop announced the completion of its seed round of financing led by First Round Capital, with participation from YCombinator and co-founders of well-known technology companies such as Instacart, Dropbox, Airtable, Mercury, and Zeus. (Newin)
Web scraping AI agent platform Reworkd receives $2.75 million in seed funding
Last year, the founder of Reworkd launched AgentGPT, a free tool for building AI agents, which became popular on GitHub and gained more than 100,000 daily users within a week. However, the co-founders soon realized that the scope of building a general AI agent was too broad, so Reworkd transformed into a web scraping company, focusing on building AI agents to extract structured data from the public network.
Reworkd has raised $2.75 million in new seed funding from Paul Graham, AI Grant, SV Angel, General Catalyst, and Panache Ventures. Together with last year’s $1.25 million pre-seed investment from Panache Ventures and Y Combinator, it brings the total raised to $4 million.
AI-driven patent application company Chekable receives NEC funding
Chekable Inc., a startup using AI to automate the patent application and search process, has closed an undisclosed investment from NEC Corp.’s Silicon Valley accelerator, NEC X.
Web3 AI Platform Assisterr Completes Seed Round of Funding
Assisterr was founded on August 25, 2023, and is committed to providing infrastructure for developers to help them build their own AI use cases. This round of financing was jointly participated by 11 well-known investment institutions such as Contango and Decasonic, and the investment amount was not disclosed.
(Welcome to add WeChat
AIyanxishe2
to learn more about AIGC and financing, and chat about new AI products with like-minded friends)
Today's big factory news
OpenAI's funding gap could reach $5 billion by 2024
The computing power OpenAI uses for AI reasoning is equivalent to 350,000 servers containing Nvidia A100 chips, of which about 290,000 are used for ChatGPT and are running at nearly full capacity. The total cost of training and reasoning AI is expected to reach $7 billion this year. As the company grows, OpenAI employee costs may reach $1.5 billion. Although OpenAI earns about $2 billion in revenue from ChatGPT each year and may earn nearly $1 billion from paid access to LLMs, it is expected to face a funding gap of $5 billion this year and may need a new round of financing. (The Information)
OpenAI tests new AI search function SearchGPT, directly challenging Google
On July 26, there was a big news in the field of artificial intelligence. OpenAI's artificial intelligence search officially began testing, directly challenging Google. OpenAI is testing a new set of search features with a limited group of users, which will answer questions with more timely information and prominent source links. This is its most direct challenge to Google, owned by Alphabet Inc., to date.
The new option, called SearchGPT, will be released as a prototype on web browsers and will offer users a standalone search experience that could later be added to its most famous product, ChatGPT. OpenAI said users will see responses to their queries with embedded attribution to creators and news publishers, including a growing number of media companies that have reached licensing agreements with the startup in recent months. OpenAI declined to disclose a timeline for the launch of SearchGPT, but said it would initially be available to users who sign up for a waitlist.
Mistral releases Large 2 flagship AI model with 123 billion parameters
Mistral's Mistral Large 2 model has 123 billion parameters. It has a 128k context window, can handle dozens of languages including Chinese, supports more than 80 programming languages, and has an accuracy of 84.0% on MMLU. Mistral pays special attention to reducing the model's hallucination problem during training. Mistral Large 2 adopts an "authorized open" strategy in terms of openness, allowing weights and fine-tuning functions to be opened for non-commercial research purposes, while commercial use requires the purchase of special licenses and usage agreements. Although the number of parameters is less than Llama 3.1's 405 billion parameters, its performance is still comparable to GPT-4, GPT-4o, and Claude 3.5 Sonnet.
Microsoft Bing launches generative AI search feature
Bing has invited some Bing users to test the AI summary feature in search results, which provides tailored dynamic response results based on the user's query. This feature places the answers generated by artificial intelligence in a prominent position on the search results page, changing the way traditional search results are arranged. Microsoft said that the Bing AI summary feature can understand search queries, review millions of information sources, dynamically match content, and generate search results with a new layout generated by artificial intelligence, thereby more effectively realizing the user's query intent.
Douyin's Youzhuju adds multiple AI and intelligent robot businesses
Tianyancha App shows that Beijing Youzhuju Network Technology Co., Ltd. has undergone industrial and commercial changes, and its business scope has been expanded to include sales of artificial intelligence hardware, research and development of intelligent robots, development of artificial intelligence application software, artificial intelligence public data platform, artificial intelligence basic software development, artificial intelligence basic resources and technology platform, artificial intelligence theory and algorithm software development, etc. The company was established in August 2017, with Xiao Jinmei as the legal representative, registered capital of US$100 million, and is wholly owned by Douyin Group (Hong Kong) Co., Ltd. (Yiou)
Stability AI launches Stable Video 4D for 3D video generation
Stability AI's latest model, Stable Video 4D, can generate eight multi-angle views of the same object from a 3D object video within 40 seconds. The model is based on the previously released Stable Video Diffusion and Stable Video 3D models, and is fine-tuned and combined with a dynamic 3D object dataset to enable it to handle object motion. The model was published on Hugging Face for developers and researchers to view and use.
????
https://huggingface.co/stabilityai/sv4d
Today Product News
Product Hunt Hot List, AI Job Search Assistant Jobright AI
Jobright AI was founded by Eric Cheng and aims to be an AI assistant for job searches, helping users discover ideal jobs, providing filtered matches, in-depth insights, and company trends. The platform integrates more than 8 million jobs, automatically updates 400,000+ new jobs every day, and filters a large number of fake jobs. Currently, it mainly serves the US market and supports job searches for H1B visas and remote work.
Co-founder Eric Cheng graduated from Zhejiang University with a bachelor's degree in computer software engineering and holds a master's degree in computer science from Carnegie Mellon University. He worked as an early full-stack engineer at Box and participated in the company's growth from zero to one hundred. After Box, he founded Fangcloud.com, a B2B content collaboration SaaS platform, which achieved $5 million in ARR and 110% ARR retention. It has received four rounds of financing from well-known venture capital firms such as Matrix Partners, Lightspeed, and JD Group. The company was acquired by Qihoo 360 Group in 2020. Currently, he is the co-founder of Jobright.ai, which aims to provide every professional with equal opportunities for self-realization through AI technology.
???? https://jobright.ai/?ref=producthunt
GitHub Trending Hot List, Enhanced Large Model Intelligent mem0
The mem0 repository is an open source project written primarily in Python that aims to provide a memory layer for large language models, enabling personalized AI interactions. The project supports multi-level memory storage, including memory retention for users, sessions, and AI agents, and can continuously improve based on interactions to provide adaptive personalization.
Mem0 provides a developer-friendly API that can be easily integrated into various applications. It also provides a worry-free hosting service solution. The project also includes the Embedchain project. Users can install it via pip. For advanced users, Mem0 supports using Qdrant as vector storage.
Future plans for Mem0 include integration with various large language model providers, support for large language model frameworks, integration with AI agent frameworks, custom memory creation/update rules, and support for hosting platforms
.
????https://github.com/mem0ai/mem0
Special attention
Research from Oxford, Cambridge and others published in Nature: Synthetic data can lead to "model collapse"
The authors explore the problems that may arise when large language models are trained using recursively generated data. As LLMs such as GPT-2, GPT-3, and GPT-4 become more and more widely used in the field of text generation, their impact on the online text and image ecosystem is also increasing.
Research has found that if the data used to train LLMs is mainly generated by previous models, then over time, the new model may forget the true distribution of the original data, leading to a "model collapse" phenomenon. This phenomenon can also be observed in variational autoencoders and Gaussian mixture models. (However, in a recent interview with Meta AI head Thomas Scialom, the head revealed that the later training of Llama 3 basically relies on pure synthetic data generated by Llama 2.)
For more details, please see:
https://www.
youtube
.com/watch?v=uLR8nSrhk_w
The researchers verified the universality of model collapse through theoretical analysis and experiments, and pointed out that this phenomenon is crucial for the long-term learning of LLMs. They emphasized that in order to maintain the benefits of LLMs, access to the original data distribution must be maintained, and more real human-generated data is needed.
????https://www.nature.com/articles/s41586-024-07566-y
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