Babbitt丨Daily must-read: Boston Dynamics uses ChatGPT and others to develop guide-guided robot dogs; OpenAI forms a new team

Boston Dynamics showcased a new study on its website that developed a talking guide-guided robot dog by combining ChatGPT, Spot, and other AI models. Smartspectrum AI today released the self-developed third-generation pedestal large model ChatGLM3, which has realized several new functional iterative upgrades, among which the multimodal understanding ability CogVLM-look graph recognition semantics has achieved SOTA effect on more than 10 international standard graphic evaluation datasets, and the semantic ability and logical ability have been greatly enhanced, which is 2-3 times faster than the current best open source implementation vLLM; OpenAI announced that it is creating a new team, Preparedness, to assess and mitigate "catastrophic risks" associated with AI. The team's main mission is to "track, assess, predict and protect" potentially significant problems caused by AI, including nuclear threats.

Image source: Generated by Unbounded AI

Hot News:

Zhipu AI released the self-developed third-generation pedestal large model ChatGLM3

reported on October 27, Zhipu AI today released the self-developed third-generation pedestal large model ChatGLM3. Zhang Peng, AICEO of Wisdom Spectrum, introduced that ChatGLM3 has achieved several iterative upgrades of new functions, among which the CogVLM of multimodal understanding capability has achieved SOTA effect on more than 10 international standard graphic evaluation datasets, and the semantic ability and logical ability have been greatly enhanced, which is 2-3 times faster than the current best open source implementation vLLM.

Gallery website Shutterstock NEW: Users can recreate existing photos through AI

As reported on October 27, gallery website Shutterstock announced the addition of a new AI-based new feature "creative editing" to improve and refine the images in the company's library of more than 700 million high-quality images.

According to reports, the feature is still in beta, taking advantage of the company's priority access to the latest OpenAI technology, users can use AI to generate new content, or simply edit and convert any image in the Shutterstock library. The "creative editing" feature, which has 6 categories, as well as a series of auxiliary functions such as AI design assistant and a series of functions such as the so-called "most advanced" filter in the industry, will be officially demonstrated live at the showcase on November 9.

Microsoft introduces a new add-on for Excel: AI to help businesses gain insights into data and predict future trends

reported on October 27, Microsoft introduced a new Office add-in for Excel Business Assist–Forecast, which can help enterprises gain insight into historical data and predict future development based on relevant data through artificial intelligence and machine learning, saving enterprises time in forecasting and analyzing data.

AI music synthesis system Loop Copilot released, generating music from multiple rounds of dialogue

As reported on October 27, the Arxiv page shows that a team of researchers from Yamaha and Queen Mary University of London and Mohammed Bin Zayed University of Artificial Intelligence recently released a music synthesis system called Loop Copilot. According to the paper, Loop Copilot enables users to generate music through iterative editing through a multi-round dialogue interface. According to the research team, Loop Copilot meets the different needs of different subsystems by using large language models to interpret user intent and selecting appropriate AI models to perform tasks. To ensure the coherence of the music, the core attributes are kept in a centralized table. Research shows that Loop Copilot has great potential to promote music creation.

Digital China Releases Big Model Platform "China Inquiry"

On October 26, Digital China Chairman and CEO Guo Wei said at the 2023 Annual Meeting of Digital China that Digital China will firmly deploy generative AI and is willing to become a partner in the whole life cycle of digital transformation of enterprises. At the meeting, Digital China released the "China Inquiry" platform, and at the same time released the industry's "Generative AI Enterprise Landing White Paper" with Xinbai Research Institute.

It is reported that the China Inquiry Platform is a one-stop enterprise big model integration platform and a large model operation platform for enterprises, which aims to help industry customers quickly build capabilities at the four levels of model, computing power, data and scenarios, and effectively open up the channel from generative AI technology to business scenarios in the enterprise.

Boston Dynamics used ChatGPT to train quadruped robots and develop a guided robot dog

As reported on October 27, robot developer Boston Dynamics showed off a new study on its official website to develop a talking guide-guided robot dog by combining ChatGPT, Spot, and other AI models. It is reported that the robot dog can talk to humans according to text and voice prompts, and provides a visual question and answer function, which can analyze the picture taken by the camera and automatically generate image descriptions.

Google Maps announces update: new AI features with support for "immersive view"

News on October 26, Google announced that the Google Maps App will usher in a series of feature updates, including new artificial intelligence features, navigation improvements and more. It is said that the new version of Google Maps will combine AI to bring more accurate search results to users. It will also display photo results based on pictures shared by other users. For example, searching for "animal latte," "find the best animal latte art," or "take the dog to the pumpkin field" will get the most immediate priority results. Google says the results are based on billions of photos analyzed by the Google Maps community. The new feature will be available this week in France, Germany, Japan, the United Kingdom, and the United States, with a gradual expansion to more countries and regions.

OpenAI is forming a new team, Preparedness, to assess the "catastrophic risks" of AI

On October 27, OpenAI announced that it was creating a new team, Preparedness, to assess and mitigate "catastrophic risks" associated with AI. The team's main mission is to "track, assess, predict and protect" potentially significant problems caused by AI, including nuclear threats. In addition, the team will work to mitigate "chemical, biological and radiological threats," as well as AI's "autonomous replication" behavior. The team is also tasked with developing and maintaining a risk-informed development policy (RDP). The RDP will detail its approach to developing rigorous cutting-edge model capability assessment and monitoring, creating a set of conservation actions, and establishing accountability and oversight governance structures throughout the development process.

Amazon CEO: Believe generative AI will generate tens of billions of dollars in revenue for AWS cloud services in the coming years

On October 27, Amazon released its third quarter 2023 financial report, with Q3 revenue of $143.08 billion, compared with $127.1 billion in the same period last year, and the market expected $141.4 billion. Amazon expects fourth-quarter net sales of $160 billion to $167 billion, versus market expectations of $166.57 billion.

Yonyou Dayi TRM based on YonGPT enterprise service model. AI2.0 is online

Yonyou Dayi TRM. AI2.0 has recently been officially announced. According to reports, Yonyou Dayi TRM. AI2.0 relies on YonGPT's large enterprise service model to implement AI technology in various recruitment application scenarios such as talent discovery, intelligent interviews, and consulting Q&A.

Honor Magic 6 will be equipped with the Snapdragon 8 Gen 3 mobile platform and support 7 billion parameter AI end-side large models

On October 25, Honor CEO Zhao Ming announced at the Qualcomm Snapdragon Technology Summit that Honor's upcoming Magic6 will be equipped with the Qualcomm Snapdragon 8 Gen3 mobile platform and support 7 billion parameters of AI end-side large models. At present, the Honor device-side AI large model can provide users with personalized services based on the understanding and perception of user preferences, combined with multi-modal natural interaction, Honor Magic6 can understand user intentions more accurately and three-dimensionally, and can also cognitively learn images, texts and complex semantics.

Xiaomi The Paper OS implants AI large models into the system and is the first to support NPU deployment

reported on October 26 that at the Xiaomi Surging OS and Xiaomi 14 series new product conference, Xiaomi MIUI head Jin Fan introduced the new Xiaomi Surging OS. It is said that Xiaomi implants AI large models into the system, and Xiaoai Input Assistant supports AI to create text, which is directly accessible with one click.

Kingsoft Office Official Wei also announced that in the research and development of new products, Xiaomi and WPS reached an in-depth cooperation, inviting users to apply to experience mobile smart office, and Xiaomi 14 series mobile phone users can download the exclusive version of WPS with WPS AI function through the Xiaomi community. In addition, Xiaomi The Paper OS also supports AI Miaohua and AI image search, users can search for images through natural language, without the need for accurate expression.

Apple and the University of Illinois release the unified vision model SAM-CLIP

According to the Arxiv page, researchers from Apple and the University of Illinois at Urbana-Champaign recently jointly published a paper introducing a unified visual basis model called SAM-CLIP. SAM-CLIP unifies the two visual foundational models of CLIP and Segment Anything Model, and combines the advantages of both models into a unified model by using technologies such as multi-task learning and continuous learning. In addition, SAM-CLIP introduces collaboration capabilities.

Investment and financing:

AI Startup Credal.ai Completes $4.8 Million in Seed Round, Led by Spark Capital

Y Combinator-backed startup Credal.ai closed a $4.8 million seed round led by Spark Capital, which aims to provide a way for businesses to connect their internal data to text-generated, cloud-hosted AI models, as reported on October 26.

** Cranium, an AI company incubated by KPMG, raised $25 million, led by Telstra Ventures**

It was reported on October 26 that Cranium, an AI company incubated by KPMG, announced a $25 million Series A financing. The round was led by Telstra Ventures with participation from KPMG LLP and SYN Ventures, bringing Cranium's total funding to $32 million.

Cranium's Connectors is a security software that helps monitor and evaluate AI usage in customer organizations in real time, supporting most major AI development environments, models, and frameworks, including Azure, Azure OpenAI, AWS Sagemaker, Google VertexAI, Databricks, MLflow, Dataiku, and DataRobot. Another product, the AI Card, launched in summer 2023, allows Cranium's customers to plug their AI applications into Cranium's secure software assessment tool and generate discrete files containing information about value, purpose, data, and governance.

Good article recommendation:

"Surrounded by giants, AI painting website survival technique"

Recently, OpenAI announced that DALL-E 3 was officially launched in ChatGPT Plus and Enterprise Edition, which means that AI painting is no longer just a fresh toy for OpenAI, but has begun to make real money.

From an industry perspective, this seems like a necessity. The higher the quality and the better the performance of AI painting, the higher the technical barriers, and the "paid wall" built on this basis has become a logical thing. But the problem is that in the current AI painting track, in addition to the Big Three, there are still a large number of open source AI paintings.

In the state of relatively weak absolute strength and open source, how do these different types of AI paintings find their own differentiation and achieve profitability?

Combining ChatGPT with physical robots, Boston Dynamics has developed a monster! 》

On October 27, Boston Dynamics, the world's top robotics developer, showed off a new study on its website that developed a talking guide-guided robot dog by combining ChatGPT, Spot, and other AI models.

It is reported that the robot dog can talk to humans according to text and voice prompts, and provides a visual question and answer function, which can analyze the picture taken by the camera and automatically generate image descriptions.

"Competing for the big model of Wenshengtu, Baidu, Kuaishou, NetEase "Battle in the open and dark"

AI painting is simply "Wen Sheng Tu", which refers to the input of some descriptive language, which AI can use to generate creative paintings. The rapid development of AIGC technology has made the "Wensheng Graph" model continue to achieve better generation results, thanks to this, whether it is an Internet giant such as Baidu and NetEase, or a cutting-edge company such as Kuaishou have rushed to enter the game, trying to explore more new possibilities in business with the help of the new thing of "Wensheng Graph".