Babbitt | Metaverse Daily Must-Read: Seven departments including the Cyberspace Administration of China jointly issued the "Interim Measures for the Management of Generative Artificial Intelligence Services", which will come into effect on August 15th. Service providers need to abide by these rules!

Abstract: Recently, the State Cyberspace Administration of China, together with the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration of Radio, Film and Television, announced the "Interim Measures for the Management of Generative Artificial Intelligence Services" (hereinafter referred to as the "Measures"). Effective from August 15, 2023. The "Measures" clarify the overall requirements for the provision and use of generative artificial intelligence services. For example, generative artificial intelligence service providers should take effective measures to prevent minor users from overly relying on or indulging in generative artificial intelligence services, mark pictures, videos and other generated content in accordance with the "Internet Information Service Deep Synthesis Management Regulations", and find illegal The content should be dealt with in a timely manner.

Image source: Generated by Unbounded AI

Hot info:

Foreign Ministry Spokesperson Wang Wenbin: Willing to strengthen communication and exchanges with the international community on artificial intelligence security governance, and promote the establishment of an international mechanism for universal participation

According to a report by the Financial Associated Press, at the regular press conference of the Ministry of Foreign Affairs of China held on the 13th, a reporter asked that Tesla CEO Musk said that China intends to establish an international cooperation framework for artificial intelligence supervision. Can China confirm and introduce the situation? "I don't understand what you said. China attaches great importance to the development and governance of artificial intelligence. It advocates adhering to the principle of people-oriented and intelligence for good, standardizes the development of artificial intelligence, and always participates in global cooperation with a highly responsible attitude." The spokesperson of the Chinese Ministry of Foreign Affairs Wang Wenbin said that China is willing to strengthen communication and exchanges with the international community on artificial intelligence security governance, promote the establishment of an international mechanism with universal participation, and form a governance framework and standards and norms with broad consensus.

HKUST Xunfei: Using the Xinghuo Cognitive Model to create a "post-diagnosis rehabilitation management platform" to generate personalized rehabilitation plans for patients

According to news from IT House on July 13, HKUST Xunfei recently stated on the interactive platform that based on the Xinghuo cognitive model, "Xunfei Medical" has comprehensively upgraded the medical post-diagnosis rehabilitation management platform, integrating "professional post-diagnosis management and rehabilitation guidance." extended beyond the yard." It is reported that the current platform can automatically analyze the patient's health profile, intelligently generate a personalized rehabilitation plan for the patient, and urge the patient to implement the plan. HKUST Xunfei said, "The pilot post-diagnosis management platform has achieved remarkable results. The platform can improve the management efficiency of doctors in cooperative hospitals by more than 10 times. The follow-up rate and consultation response rate during the recovery process of patients reached 100%. More than 98%".

Google's large medical model is listed in Nature, and its level is comparable to that of clinicians

According to the "Kechuangban Daily" report, according to Google's paper published in "Nature" on July 12, when answering medical questions, its fine-tuned medical large model Med-PaLM performed well, and a group of clinicians answered it. The score was 92.6%, which was comparable to the level of clinicians in reality (92.9%). In addition, Google proposed a new MultiMedQA evaluation benchmark, covering questions and answers in medical examinations, medical research and other fields, to evaluate the clinical capabilities of large models.

Jingdong Yanxi large model plans to open capacity in the first half of 2024

According to Jinshi’s report on July 13, JD’s 100-billion-level Yanxi model was unveiled today, supporting content generation, man-machine dialogue, and understanding of user intentions. He Xiaodong, dean of Jingdong Exploration Research Institute, introduced that the Yanxi large model supports language, voice, vision, multi-modality, etc., 70% of general domain data, combined with 30% of industry data, to achieve fine-tuning of the base model + vertical domain model, It can serve retail, finance, education, government affairs and other fields. He Xiaodong revealed that it plans to fully open up the large-scale model capability in the first half of 2024 and export customized models to the industry. In addition, JD.com is also exploring the transition from a large language model to multi-modal digital human interaction, and embodied intelligence will also be the key exploration direction of JD.com Research Institute in the future. As previously reported, the JD Yanxi model is expected to be officially launched in August.

Chinese Online: Reached cooperation with a number of large model companies in terms of data, and the contract value of a single company has reached the level of tens of millions of yuan

According to Jiemian News, Chinese Online said on the investor interaction platform on July 12 that the company has recently reached cooperation with a number of large-scale model companies on artificial intelligence large-scale model data, and the contract value of a single company has reached tens of millions of yuan. The company will provide it with genuine data content resources to support its research and development and training of artificial intelligence large model technology and applications. At present, the company has more than 60 TB of text, audio, and video data, and can provide customized, multi-type, and cross-modal large-model data solutions, which are widely used in the field of AI +.

US venture capital firm Sapphire Ventures plans to invest more than $1 billion in AI startups

According to a report by the "Technology Board Daily" on July 13, Jai Das, president of the venture capital firm Sapphire Ventures, announced at a meeting on Tuesday that the company plans to invest more than US$1 billion in artificial intelligence startups to accelerate the adoption of the technology. Adoption in global enterprises. "As AI is affecting how new software is built, it's really changing the game," Das said. AI startups have raised more than $40 billion in funding over the past six months, according to research firm PitchBook. Accounting for nearly a quarter of total funding for start-ups. It is reported that Sapphire Ventures is a venture capital institution with an asset management scale of over US$4 billion.

Google's AI chatbot Bard auditors say they are under pressure and only have 3 minutes to review answers

According to IT House citing a Bloomberg report on July 12, Google's AI chatbot Bard was trained under pressure by thousands of contract workers (Contractor), who only had 3 minutes to review the answers generated by Bard. According to reports, these contract workers come from companies such as Appen and Accenture. They have received only minimal training and are paid as little as $14 an hour (about 100 yuan). Their job is to assess whether Bard's answers are reliable and accurate. "As things stand, people are scared, stressed, underpaid and don't know what's going on. This culture of fear is not conducive to our quality and teamwork," said a contractor.

In response, a Google spokesperson said in a statement: "Giving people access to high-quality information is at the core of our mission. We have done a lot of work to responsibly build our artificial intelligence products, including our continuous improvement over the years. Our rigorous testing, training, and feedback process to emphasize factuality and reduce bias. Human evaluations, including individuals from inside and outside of Google, are one of the many ways we use to improve our products.” The spokesperson also said, “Scoring do not directly affect the output of our models, and they are by no means the only way we can improve accuracy."

People familiar with the matter: Meta will release commercial versions of AI models, allowing companies to build custom software

Meta is preparing to release a commercial version of its AI model, allowing startups and businesses to build custom software on top of it, the Financial Times reported, citing sources. The move will allow it to compete with Microsoft-backed OpenAI and Google for a leadership position in generative artificial intelligence development. The software, which currently creates text, images and code, is powered by large-scale language models (LLMs) that are trained on vast amounts of data and require massive computing resources. Meta released the language model LLaMA to researchers and academics earlier this year, and the new version will be made more widely available and customizable by businesses, and a release of that version is expected soon, the people said.

Glory Zhao Ming: Folding screens and end-to-side AI models are the future direction of innovation

According to a report by "Kechuangban Daily" on July 12, talking about the future direction of innovation in the field of consumer electronics, Zhao Ming, CEO of Honor, said in an interview with the media that, first of all, folding screens are a visible opportunity in the next 3-5 years. It is the end-to-end AI model. In addition, the combination of multiple devices and all scenarios to meet the needs of consumers is also the future development direction.

Investment and Financing:

artificial intelligence-based large-scale model company reInvent AI completed tens of millions of dollars in the first round of financing, Lenovo Venture Capital exclusive investment

According to 36 Krypton’s report on July 13, recently, reInvent AI announced that it has completed tens of millions of dollars in the first round of financing, which was exclusively invested by Lenovo Venture Capital. reInvent AI focuses on basic models and general artificial intelligence (AGI), aiming to create industry-oriented professional large models. The founding team members include academicians of the European Academy of Sciences, core members of OpenAI GPT, and the former SAP global senior vice president and global AI leader, as well as the original author of the Transformer paper as a consultant. reInven tAI will operate simultaneously in Berlin, Germany and Shanghai, China, and its German company is called nyonic.

Metaverse recommended reading:

"Interim Measures for the Management of Generative Artificial Intelligence Services" jointly announced by the State Cyberspace Administration of China and other seven departments

Recently, the State Cyberspace Administration of China, together with the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration of Radio, Film and Television, announced the "Interim Measures for the Administration of Generative Artificial Intelligence Services" (hereinafter referred to as the "Measures"). It will come into force on August 15th. The relevant person in charge of the State Internet Information Office stated that the introduction of the "Measures" aims to promote the healthy development and standardized application of generative artificial intelligence, safeguard national security and social public interests, and protect the legitimate rights and interests of citizens, legal persons and other organizations.

"Musk formed xAI company to "understand the true nature of the universe", with members from well-known AI institutions"

Tesla founder Elon Musk's artificial intelligence company finally has definite information. On July 12, local time, Musk’s newly established artificial intelligence company xAI launched a new website (detailing its mission and team. Musk said on Twitter that the company’s purpose is to “understand reality.” The website says : "The goal of xAI is to understand the true nature of the universe."

"GPT-4 "Alchemy" Guide: MoE, Parameter Quantity, Training Cost and Reasoning Secrets"

GPT-4 is the result of a deep combination of science and engineering innovation. There are countless tricks in the middle. For the outside world, if you can understand the structure of GPT-4, it is like obtaining the "recipe for alchemy" of the strongest model. This content gives the GPT-4 architecture, training and inference infrastructure, parameter quantity, training data set, token number, cost, and MoE model and other parameters and information details in great detail.