OpenAI and Thrive Capital invest in six-month-old AI drug discovery company Chai Discovery
Chai Discovery, a six-month-old AI biotech startup, has raised nearly $30 million from heavyweight venture capital firms Thrive Capital and OpenAI to use AI for drug discovery. The company's basic model is designed to compete with Google DeepMind. It has just released the open source model Chai-1 for molecular structure prediction and a technical report. This round of financing was led by Thrive and followed by Dimension Capital, and the company is valued at $150 million.
Prior to joining Chai, co-founder Meier served as chief AI officer at biotech company Absci Corp. and worked as a researcher at Meta and OpenAI. He co-founded the startup in March with Jack Dent, who worked as an engineer at Stripe Inc., among others. Chai has fewer than 10 employees so far, but it has attracted talent from companies including OpenAI, Google and Meta, Meier said.
Chai-1 Technical Report
????
https://www.chaidiscovery.com/blog/introducing-chai-1
AI search startup Glean raises $260 million at $4.6 billion valuation
Artificial intelligence search startup Glean said Tuesday it has raised $260 million in a funding round that values it at $4.6 billion, more than double its last reported valuation. The
Palo Alto, California-based company, which ranked No. 43 on this year’s CNBC Disruptor 50 list, has raised more than $600 million to date from more than 20 investors.
Glean competes with a host of well-funded generative AI startups and tech giants, trying to take on Microsoft Copilot and chatbot Amazon Q. It also aims to disrupt the field of cognitive search tool providers such as Perplexity, Coveo, Sinequa and Lucidworks.
Final, a startup focused on accounting and finance for small and medium-sized businesses, raises $200 million in equity and debt funding
Finally offers AI-driven bookkeeping services, corporate cards, AI ledgers, and payroll processing, among other services, aiming to be an all-in-one financial management platform for small and medium-sized businesses. Final raised $50 million in Series B funding and secured a $150 million credit line. PeakSpan provided the equity portion of the financing, while Encina provided a $150 million credit line.
SafetyCulture, the AI-driven enterprise operations platform, raises £85 million in investment
SafetyCulture's platform supports 85,000 companies and nearly 2 million users worldwide, with capabilities that enable businesses to manage safety, quality and operational inspections. Its latest advancements in artificial intelligence are designed to transform the way frontline teams interact with data, improving workplace safety and boosting productivity.
The round, led by Airtree Ventures with support from existing investors Blackbird and Morpheus Ventures, comes shortly after the company achieved unicorn status at a valuation of £1.3bn.
Robotics Company Richtech Robotics Announces Completion of $21 Million Public Offering
Richtech Robotics, a provider of AI-driven service robots, raised $21 million through a public offering of Class B common stock and warrants, with the potential to raise an additional $21 million if all warrants are exercised. Rodman & Renshaw LLC served as the exclusive placement agent for the offering.
PIN AI raises $10 million in pre-seed funding to advance decentralized AI platform
PIN AI, a participant in venture capital firm Andreessen Horowitz's Fall 2024 a16z crypto startup accelerator, is building a decentralized on-device AI platform that prioritizes user privacy and data sovereignty. The platform is open source and allows developers to create custom AI agents that can securely interact with user data, giving users full ownership of their information. Compared to Apple Intelligence, its decentralized model provides flexibility for developers and users.
PIN AI’s pre-seed investors include a16z CSX, Hack VC, Anagram Ventures LP, Symbolic Capital Partners LP, DCbuilder Inc., Foresight Ventures Ltd., Blockchain Builders Fund LP, Alumni Ventures Group and some individual investors.
AI agreement platform
Agree.com
raises $3 million
Agree has developed an AI-powered infrastructure that automatically extracts key information from contracts, such as name, title, address, and payment terms, and can generate and issue invoices. The platform also allows users to collaborate with multiple parties in real time when editing contracts.
This round of financing was led by Better Tomorrow Ventures, with participation from 8-Bit Capital, Everywhere Ventures, Expedite Ventures, Firsthand Alliance, Hustle Fund, NEA, Singh Capital Partners and Trust Fund.
(Welcome to add WeChat
AIyanxishe2
to learn more about AIGC and financing, and chat about new AI products with like-minded friends)
Sichuan Embodied Humanoid Robot Technology Company was established
Tianyancha App shows that Sichuan Embodied Humanoid Robot Technology Co., Ltd. was established, with Feng Zhenyu as the legal representative and a registered capital of RMB 20 million. Shareholder information shows that the company is jointly held by Chengdu Xinzhiyingxing Technology Partnership (Limited Partnership), Chengdu Zhongqi Technology Partnership (Limited Partnership), Sichuan Machinery Research and Design Institute (Group) Co., Ltd., and Chengdu Chuanhagong Robot and Intelligent Equipment Industry Technology Research Institute Co., Ltd., and the beneficial owner is Peng Bei.
It is reported that Sichuan Embodied Humanoid Robot Technology Co., Ltd. is an emerging enterprise built by concentrating the province's advantageous resources. It is led by Professor Peng Bei, vice president of the Sichuan Artificial Intelligence Industry Association and director of the Sichuan Intelligent Service Robot Engineering Center, and plans to launch its first humanoid robot before the end of the year.
AMD CEO Lisa Su said at the Communications and Technology Conference held by Goldman Sachs that the development of artificial intelligence has exceeded her expectations five years ago, and AMD is making major investments for the next five years. AMD plans to launch the MI325 AI chip later this year, followed by the MI350 and MI400 chips in 2025 and 2026 respectively. Su Lisa revealed that the new generation of MI series chips will have up to 192GB of memory and 153 billion transistors of processing power, which can be used to train large language models such as OpenAI's ChatGPT.
Apple releases the first AI iPhone
Apple unveiled the iPhone 16 series, the first native AI phone in its history. The highlight of the new series is the new Apple Intelligence, which combines the power of personal context and generative models to provide a more personalized and privacy-preserving AI experience. The iPhone 16 series is equipped with A18 and A18 Pro chips, which use a 3nm manufacturing process to significantly improve performance and energy efficiency. Apple Intelligence will be launched in Beta next month, and more features will be released in the following months. The Chinese version is expected to be launched next year.
Salesforce launches Industries AI platform to create autonomous agents for industries
Salesforce has launched Industries AI, which aims to help companies efficiently solve various industry challenges by creating industry-specific autonomous agents. The platform will be embedded in all 15 of Salesforce's cloud platforms to handle more than 100 common tasks. Industries AI's solution is to automatically create artificial intelligence agents by providing a ready-made framework that leverages each customer's proprietary data within the Salesforce platform. Salesforce has also created a dedicated use case library for each of its cloud platforms to help customers get started creating artificial intelligence automation. These features span multiple industries.
Transformer converted to linear RNN, Together AI increases inference speed by 1.6 times
Together AI's latest research combines the LIama 3 model with the Mamba model, and converts the Transformer model into a linear RNN structure through a special distillation method. Experimental results show that the hybrid model performs well in different NLP tasks, with an inference speed increase of more than 1.6 times, and its performance on the OpenLLM Leaderboard list is comparable to or even better than the best open source RNN model.
????
https://www.together.ai/blog/the-mamba-in-the-llama-distilling-and-accelerating-hybrid-models
Nous Research's distributed optimizer DisTrO reduces LLM training communication by 10,000 times
Nous Research announced a major breakthrough in the distributed optimizer DisTrO, which can significantly reduce the amount of communication between GPUs when training large language models, from 1,000 to 10,000 times. This progress alleviates one of the important costs and bottlenecks of large model training, bandwidth. In the experiment, Nanotron was used as the pre-training framework, and the 1.2B size Llama 2 was selected for testing. The results showed that DisTrO reduced the communication volume from 74.4GB to 86.8MB, which is equivalent to a 857-fold reduction in bandwidth pressure.
????
https://github.com/NousResearch/DisTrO/blob/main/A_Preliminary_Report_on_DisTrO.pdf
Peking University and Baichuan Intelligence jointly developed the automatic prompt engineering system PAS
Peking University-Baichuan Joint Laboratory has developed an automatic prompt engineering system PAS, which is based on a large language model with a Transformer architecture and aims to improve the performance of the model on specific tasks. The researchers conducted a comprehensive evaluation of PAS through three benchmarks and applied it to six top AI models. The evaluation results show that PAS has achieved significant performance improvements and demonstrated compatibility and adaptability with various ultra-large models. In addition, PAS also excels in computational efficiency, with only 9,000 fine-tuning data required to demonstrate excellent performance, and can limit the length of supplementary automatic prompts to improve user experience.
???? https://github.com/PKU-Baichuan-MLSystemLab
Product Hunt
hot list, SciSpace AI Writer
SciSpace AI Writer is an AI paper writing tool launched by SciSpace, following SciSpace Copilot which became the third most popular product the day before. It is designed to solve the difficult task of writing academic documents through AI.
The tool leverages training data from over 280 million research papers and serves different user groups, including students, scholars, and researchers. Users can enjoy the following features: select and adjust the tone of text with one click, export output to over 75 languages, quickly create outlines with slash commands, get fast auto-completion help, write freely with the rich text editor, cite or add your own sources from over 280 million research papers.
In addition, SciSpace AI Writer allows users to save notes to drafts anywhere on the platform, making the research and writing process more seamless. Currently, the SciSpace platform has nearly 2 million users worldwide.
???? https://typeset.io/ai-writer?ref=producthunt
1. Reshot AI, a facial expression editing tool
Reshot AI is a professional AI photo editor for editing and enhancing personal photos, such as social media avatars and YouTube thumbnails. It provides multiple editing options through AI technology, including adjusting expressions, facial poses, lighting, and backgrounds. Users can generate multiple optimized photo options by uploading a photo and selecting the desired editing options.
???? https://www.reshot.ai/
2.01 App, making your phone a universal remote control
Open Interpreter has launched the free application 01 App, which can be used on iOS and Android devices. It integrates advanced speech recognition and generation technology, allowing users to control their Mac, Windows and Linux machines through a simple push-to-talk interface. Users can remotely control their computers through 01 App from anywhere in the world. The application is able to accumulate contextual information when the button is pressed, and can continue to process and understand the user's voice commands even if the button is not pressed.
Open Interpreter will also open source its entire manufacturing process, including product requirements documents, electrical diagrams, firmware architecture, and industrial design.
????https://changes.openinterpreter.com/log/01-app
Can LLMs generate novel research ideas?
In a recent Stanford study, researchers designed a one-year experiment to explore whether large language models can generate research ideas with expert-level and novelty. During the experiment, a total of 49 expert researchers in the field of natural language processing were invited to write novel ideas on 7 NLP topics. At the same time, the research team built an LLM agent to generate research ideas on the same 7 topics as human experts. Afterwards, 79 experts were recruited for blind review to evaluate these ideas generated by humans and LLMs.
The results showed that the ideas generated by LLM surpassed those of human experts in terms of novelty. The participants included PhDs and postdocs from 36 different institutions, and the number of citations of their research results indicated that they had high academic status in the field. To ensure the completeness and feasibility of the ideas, the research team provided detailed idea templates and paid $300 for each idea, with an additional $1,000 bonus for the top five human ideas.
???? https://x.com/ChengleiSi/status/1833166031134806330
????https://arxiv.org/abs/2409.04109
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.