Babbitt | Metaverse Daily Must-Read: ByteDance launched the AI dialogue product "Doubao", followed by Kuaishou announced that the Android version of "Kaishou AI dialogue" is open for internal testing, domestic large-scale model melee is still hot

Abstract: According to 36 Krypton reports, on August 17, ByteDance announced the start of external testing of the AI dialogue product "Doubao", which can be used by entering the official website to register directly or download the app through the QR code provided by the official website. However, the beanbags thrown by Byte were still hot, and Byte’s old opponent, Kuaishou, obviously couldn’t hold back. On the evening of August 18, Kuaishou suddenly announced that the Android version of "Kuaishou AI Dialogue" was open for internal testing. Previously, Kuaishou AI had conducted a small-scale test in early August, and Kuaishou AI Dialogue was directly built into Kuaishou. "Shaking Kuai" has released AI applications, which is undoubtedly another important node for the landing of large-scale models in China.

Image source: Generated by Unbounded AI

Hot info:

Baidu will release a variety of AI native applications on October 17

According to a report from IT House on August 21, Baidu recently announced that on October 17, Baidu World 2023 with the theme of "Generating the Future (THE WORLD)" will be held in Beijing Shougang Park. Baidu will release a variety of AI-native applications, introduce the latest developments in large models, and share how to seize new opportunities with enterprises, users, and partners through AI-native thinking.

The conference will also hold sub-forums for enterprises, developers, partners, investors, etc., and set up a cutting-edge AI interactive experience area to fully demonstrate Baidu's latest achievements. The data shows that Wenxin Yiyan has mastered more than 200 creative genres since its release in March this year. The richness of content is 1.6 times that of the initial release, the length of the thinking chain is 2.1 times that of the initial release, and the coverage of knowledge points is 8.3 times that of the initial release. .

Australian Department of Defense staff used ChatGPT thousands of times without approval

According to the webmaster's home report on August 21, according to Crikey's freedom of information request documents, Australian Ministry of Defense staff used the artificial intelligence chatbot ChatGPT thousands of times without approval. The documents also suggest that the country's Ministry of Defense has restricted access to the web domain of ChatGPT owner OpenAI to prevent data or privacy breaches.

It is alleged that between December 1, 2022, and June 30, 2023, Australian Department of Defense equipment connected to the OpenAI.com web page 5,630 times. These connections are made through an online web interface, excluding connections made through third-party services or OpenAI's smartphone application.

David Evans, the country's Ministry of Defense designated decision maker, provided additional information on the Ministry of Defense's ChatGPT policy in an accompanying letter. He wrote that the Department of Defense has not approved access to OpenAI's products and has restricted the use of "online AI services such as ChatGPT" on defense equipment. Currently, there is no government-wide guidance on the use of generative artificial intelligence products such as ChatGPT by Australian federal departments.

Beijing will strengthen the supervision of Internet diagnosis and treatment: it is strictly forbidden to use AI to automatically generate prescriptions

According to a report from Beijing Daily on August 21, the Beijing Municipal Health and Health Commission took the lead in formulating the "Implementation Measures for the Supervision of Internet Diagnosis and Treatment in Beijing (Trial)" and publicly solicited opinions from the public. committee feedback. The method proposes that medical institutions must strengthen drug management when carrying out Internet diagnosis and treatment activities, and it is strictly forbidden to use artificial intelligence to automatically generate prescriptions, and it is strictly forbidden to provide drugs to patients before prescriptions are issued.

The supply chain reveals that Dell will stock 20,000 AI servers this year

According to the "Kechuangban Daily" citing Taiwan's Economic Daily, according to Dell's latest AI server bill of materials (BOM) and stocking plan revealed by the supply chain, Dell's AI server stocking volume this year looks at 20,000 units, and it will increase again next year. Dell is aggressively attacking AI servers, sweeping up Nvidia H100 chips, and releasing a large number of OEM and component orders. The total order value this year is estimated to be as high as 100 billion yuan, and it is expected to double to more than 200 billion yuan next year.

AI2 released the open data set Dolma, breaking the data barrier of AI language model

According to a report from the Webmaster's Home on August 21, the Allen Institute for Artificial Intelligence (AI2) released an open text dataset called Dolma, which aims to promote transparency and innovation in AI language models. Dolma, as the core of the AI2 Open Language Model (OLMo) initiative, will provide researchers and developers with free data resources to support broader AI research.

Dolma serves as the cornerstone of the AI2 Open Language Model (OLMo) initiative, and its acronym stands for "Data to feed OLMo's Appetite" (data to feed OLMo's appetite). AI2 aims to encourage more innovation and research by providing free, open data resources to the research community.

Meta plans to launch open source AI programming software Code Llama this week

According to a report from the Webmaster's House on August 21, Meta plans to launch the open source AI software Code Llama this week, which aims to help developers automatically generate programming code. This groundbreaking product will challenge the proprietary software of tech giants such as OpenAI and Google. Additionally, Meta is developing an AI-assisted code-writing tool called CodeCompose. The tool has expanded to Meta's tens of thousands of developers across multiple programming languages and coding interfaces.

Kuaishou's self-developed large model "Ruyi" has started internal testing, and the new version ranks first in CMMLU Chinese

According to a report by Webmaster's Home on August 21, the large-scale model "Ruyi" independently developed by Kuaishou performed well on several large-scale model lists. Among them, the new version of KwaiYii-13B ranked first in CMMLU Chinese. At present, "Ruyi" has started internal testing, and provides API and customized project cooperation solutions for business teams.

It is reported that the "KwaiYii" large model (KwaiYii) is a series of large-scale language models (Large Language Model, LLM) independently developed by the Kuaishou AI team from zero to one. Training model (KwaiYii-Base), dialogue model (KwaiYii-Chat).

Among them, the KwaiYii-13B-Chat dialogue model has excellent language understanding and generation capabilities, and supports a wide range of tasks such as content creation, information consultation, mathematical logic, code writing, and multiple rounds of dialogue. The manual evaluation results show that KwaiYii-13B-Chat exceeds the mainstream The open source model is close to the same level as ChatGPT (3.5) in content creation, information consultation and mathematical problem solving.

Investment and Financing:

AI technology company "Biyou Technology" completed the seed round of financing, and the funds will be used for product development, etc.

According to a report by the "Kechuang Board Daily" on August 21, Biyou Technology has recently completed a seed round of financing, but did not disclose the amount of the financing and the investors. Wen Kaiyu, CEO of Biyou Technology, said that the new funds will be used for product research and development, talent introduction and market construction. Biyou Technology is an AI technology company that focuses on the research and development of cutting-edge artificial intelligence technologies and products. Its biU AI is the world's first revolutionary AI product for front-end insurance salespersons.

Korea SK Telecom announces strategic investment in AICC developer Persona AI

According to Jiemian News, SK Telecom, the largest telecom operator in South Korea, announced on August 21 that it will make a strategic investment in Persona AI, a South Korean Intelligent Contact Center (AICC) developer, and become the third largest shareholder of the latter. SK Telecom plans to strengthen its AICC division to become a comprehensive provider of AI contact center processing services including consulting, infrastructure construction and maintenance. Persona AI has a self-developed natural language processing engine and launched Korea's first subscription-based AICC service.

Metaverse recommended reading:

"Byte Online Doubao App: Xueba, Chat Partner and Specialist Teacher, Fighting with Kuaishou AI"

Byte, which has been silent for half a year, finally released the latest progress of its own large model! There is no signal. On August 17, ByteDance announced the start of external testing of the AI dialogue product "Doubao", which can be used by entering the official website to register directly or download the app through the QR code provided by the official website. However, the beanbags thrown by Byte were still hot, and Byte’s old opponent, Kuaishou, obviously couldn’t hold back. On the evening of August 18, Kuaishou suddenly announced that the Android version of "Kuaishou AI Dialogue" was open for internal testing. Before that, Kuaishou AI had conducted a wave of small-scale testing in early August. But Byte Doubao is still an independent app, but it is clear that Kuaishou wants to take the lead in the application, and the Kuaishou AI dialogue is directly built into Kuaishou.

"The open source data set used by LLaMA was taken off the shelf: it contains nearly 200,000 books, and it is benchmarked against the OpenAI data set"

The open source data set was removed from the shelves due to copyright infringement. Such as LLaMA, GPT-J, etc., have been trained with it. Today, the website that hosted it for 3 years deleted all related content overnight. This is Books3, a data set consisting of nearly 200,000 books, with a size of nearly 37GB. A Danish anti-piracy organization stated that 150 books of its members were found in the data set, which constituted infringement, so it asked the platform to remove it. Now the Books3 web page link on the platform has been "404". The original developer of the data set said helplessly that the removal of Books3 is a tragedy in the open source circle.

"Nvidia and cloud giants must have a battle"

In 2023, Nvidia on the run encountered generative AI, which Huang Renxun called "AI's iPhone moment" many times. At this moment, Huang Renxun set his sights on an opportunity to become the hunter rather than the prey - Yun.