ByteDance Launches Flow, a new AI division, and Nvidia sold nearly 500,000 A100 and H100 AI GPUs in Q3

Summary: ByteDance recently established a new AI division, Flow, which focuses on the AI application layer. Thanks to the hot demand for AI GPUs, Nvidia made $14.5 billion on data center hardware. Nvidia sold nearly 500,000 A100 and H100 GPUs in Q3, and lead times for H100-based servers have been extended to 36 to 52 weeks, according to research firm Omdia.

Image source: Generated by Unbounded AI

Hot News:

ByteDance established a new department, Flow, to focus on the AI application layer

According to a report by 36Kr on November 27, ByteDance recently established a new AI department, Flow, and the technical leader is Hong Dingkun, vice president of technology at ByteDance. A person familiar with the matter said that the business leader of this new department is Zhu Wenjia, the head of the byte model team.

It is reported that Flow mainly focuses on the AI application layer. In the byte circle, Flow recently released a living water recruitment post, and social recruitment has also started for a while. In the post, it said that it is ByteDance's AI innovation business team, "At present, two products have been launched in China and overseas, Doubao and Cici, respectively, and a number of AI-related innovative products are being incubated." As of press time, ByteDance has not responded.

Omdia: Nvidia Q3 sold nearly 500,000 A100 and H100 AI GPUs, Meta and Microsoft are the biggest buyers

According to IT Home on November 28, in the third quarter of this fiscal year, thanks to the hot demand for AI GPUs, Nvidia earned $14.5 billion on data center hardware. Nvidia sold nearly 500,000 A100 and H100 GPUs in Q3, and lead times for H100-based servers have been extended to 36 to 52 weeks, according to research firm Omdia. According to Omdia, Meta and Microsoft are the biggest buyers of Nvidia's H100 GPUs. They all procured up to 150,000 H100 GPUs, far more than Google, Amazon, Oracle, and Tencent bought the number of H100 processors (50,000).

Hugging Face Co-Creates Releases 2024 Prediction: Open Source LLMs Will Reach the Best Closed-Source LLM Levels

AI New Intelligence News, on November 27, Clement Delangue, co-founder and CEO of Hugging Face, an AI open source community, posted 6 predictions for the development of the industry in 2024:1. A popular AI company will go out of business or be acquired at a very low price;2. Open source LLMs will be at the best level of closed-source LLMs;3. AI is making major breakthroughs in video, time series, biology and chemistry;4. The public will become more concerned about the economic and environmental costs of AI;5. Most of the content of a popular media outlet will be generated by AI;6. 10 million AI developers on Hugging Face will not lead to an increase in unemployment.

Google rolls out AI Core update on Pixel 8 Pro with latest AI model support

According to the webmaster's home on November 28, Google today launched a new Google app update called AI Core. According to the Google Play Store, AI Core will provide the latest AI model support for the Google Pixel 8 Pro phone's Android system and apps, as well as a range of other AI-powered features.

China Telecom 500 million to establish a digital technology company, including AI software development business

According to a report by the Financial Associated Press on November 28, Tianyancha showed that recently, China Telecom Digital Government Technology Co., Ltd. was established, the legal representative is Cai Xudong, with a registered capital of 500 million yuan, and its business scope includes software development, big data services, artificial intelligence basic software development, cloud computing equipment technical services, network technology services, electronic product sales, Internet access services, etc. Shareholder information shows that the company is wholly owned by China Telecom.

Microsoft Copilot "Brain" will be upgraded to GPT-4 Turbo and the character limit will be raised

According to IT Home on November 28, Mikhail Parakhan, head of Microsoft's advertising and network services, recently tweeted on the X (Twitter) platform, saying that the team is actively improving Copilot (Bing Chat), and after solving some problems, it will be upgraded to OpenAI's GPT-4 Turbo. Parakhin also said that after upgrading GPT-4-turbo, the character limit will be further expanded, and the maximum number of words can now be entered is 4,000 words.

Microsoft has introduced DALL-E 3 integration for Win11 Paint app

According to IT Home on November 28, Microsoft has now begun to push the "Paint" app integrated with DALL-E 3 to all Windows 11 users, which can be opened through the "Cocreator" button. However, the feature was previously limited to a select group of users in the US, France, UK, Australia, Canada, Italy, and Germany, and it is unclear if it will be available to all users.

The DALL-E model is an image generation model developed by OpenAI that has previously been used in Microsoft's Bing image generation feature.

Taotian Group is preparing to build a large model team, with Dai Shan, Ruohai and Zheng Bo in charge

According to a report by Lei Feng on November 27, Taotian Group is preparing to build a large-scale model research team and has started recruitment. It is reported that the large-scale model research of Taotian Group will mainly focus on two scenarios: one is to search and push, and the other is to browse the content. The team building was led by Dai Shan, CEO of Taotian Group, Ruohai, CTO of Taotian Group, Zheng Bo, CTO of Alimama, and others. A technical talent related to the large model said that the large model position of Taotian Group is still very attractive. On the one hand, Taotian Group has a very strong ability to resist risks in the current environment, and on the other hand, the application scenarios of Taotian Group's large model are also relatively clear, which is more certain than that of a general large model.

Meta's new study enables large models to automatically mask irrelevant information, increasing accuracy by 27%

As reported by Qubit on November 27, Meta recently unveiled a new study on the mechanism of attention. The research team named this new attention mechanism "2 Attention" (S2A), which can block the interference of irrelevant information and further improve the accuracy of large models. According to the test results, the mechanism does not require fine-tuning or training, and can increase the accuracy of large models by 27% on its own.

Investment & Financing:

Biotech and AI startup Cradle radle raises $24 million in Series A funding

Biotech and artificial intelligence startup Cradle has raised $24 million in Series A funding after closing a $5.5 million seed round last year, following a $24 million Series A funding round led by investor Index Ventures and with participation from Kindred Capital (also a seed round investor) and individual investors such as Chris Gibson and Tom Glocer, with the new funds to be used to grow the team and sales, TechCrunch reported on November 28.

It is reported that Cradle's technology can greatly shorten the manufacturing time of functional proteins and reduce the number of experiments. Currently, the company has produced an alternative version of T7 RNA polymerase, an RNA-producing enzyme, using its own software.

AIGC荐读:

Hugging Face CEO Prediction: Six Great Changes in the AI Industry in 2024!

OpenAI Chairman Greg Brockman predicted on the last day of last year that 2023 would make 2022 look like AI development has not yet woken up. Sure enough, in 2023, the AI industry will usher in a full-scale explosion. Yesterday, Clement Delangue, CEO of Hugging Face, the world's largest AI open source community, made 6 specific predictions for the development of the industry in 2024:

"Meitu's AI story is still "old wine in new bottles"

If you count from the official release of chatGPT, in less than a year, this storm of AI large models sweeping the world is subverting everything, including Chinese companies at the moment. In less than a year, Chinese companies have scrambled to release hundreds of industry models, and in various evaluation lists and articles, "the most capable" seems to have become the official unified term. Although, some consensus has been determined in many analysis articles, such as the current domestic large model boom has calmed down, and for example, the competition pattern of basic large models has been determined, and the wrestling of the application layer has begun. But if we leave the excited context of geeks and return to a company and an industry itself, we may have a more pragmatic perspective and solve some more pragmatic and direct questions: what is a company's AI strategy, how will it affect the future of the company, and is it the one who is the one who makes the trend?

Huang Jenxun bets on artificial intelligence to build a trillion-dollar graphics card empire**

As the soul of the company, what story has Huang experienced along the way?