AI Era Network Comprehensive Report丨【WeChat Subscription:AI Era Network】
AI Era Network June 27, 2023 morning report
Robin Li announced the upgrade of Wenxin Big Model V3.5: the inference speed has been increased by 17 times
At the Nishan Forum of the World Internet Conference on June 26, Baidu CEO Robin Li announced that Baidu Wenxin big model has been iterated to version 3.5, compared with version 3.0, the training speed has been increased by 2 times, the inference speed has been increased by 17 times, and the model effect has been improved by more than 50%.
Wenxin model version 3.5 is not only a technical upgrade, but also a security upgrade. The results show that version 3.5 of the Wenxin model has been significantly improved in terms of data quality, generation effect and content security.
On July 7, Huawei announced an important upgrade to the Pangu model: programming acceleration
Huawei HDC 2023 Developer Conference will be held in Dongguan on July 7. At present, Huawei has announced the keynote speech agenda of this conference, and AI occupies a lot of content, especially the HUAWEI CLOUD Pangu model will usher in a major upgrade. Huawei's Pangu series of basic large models were officially released in 2021, including NLP (natural language processing), CV (machine vision), and scientific computing big models.
It is understood that HUAWEI CLOUD CodeArts Snap intelligent programming assistant is an application case of Huawei's large code model, which can empower developers to develop efficiently and trusted. This means that Huawei's large-model applications are accelerating to the global market and continue to expand the ecosystem.
Zhou Hongyi: Many enterprises only need tens of billions of large models, and hundreds of billions is waste
On June 26, at the main forum of the Nishan Dialogue on Digital Civilization at the World Internet Conference, Zhou Hongyi, founder of 360 Group, shared relevant practical experience and thinking on 360's self-developed cognitive large model "Wisdom Brain" as an example. He said: "The birth of the GPT large model represents the arrival of general artificial intelligence and strong artificial intelligence, and it is the emergence of real intelligence. ”
First, although the public-owned model is a generalist, it lacks industry depth. Second, the public large model is easy to cause internal data leakage. Third, for enterprises, the public model cannot guarantee that the content is truly credible. Fourth, the public large model cannot achieve controllable costs.
At the same time, many enterprises actually only need the ability to write code with large models, tens of billions of vertical large models can meet the demand, if you use hundreds of billions of large models is a waste of cost. In terms of controlling costs, vertical large models will have great advantages. On the basis of the "general" big model trained with open data, training a proprietary large model can achieve "double the effort with half the effort" and reduce costs and increase efficiency for enterprises.
Tencent's self-developed Xingmai high-performance computing network: helps AI models improve communication performance by 10 times
Tencent Cloud, a subsidiary of Tencent, fully disclosed its self-developed Xingmai high-performance computing network for the first time. It is said that Xingmai Network has the industry's highest 3.2T communication bandwidth, which can improve GPU utilization by 40%, save 30%~60% of model training costs, and bring 10 times the communication performance improvement to AI large models. Based on Tencent Cloud's next-generation computing power cluster HCC, it can support a large computing scale of 100,000 cards.
Wang Yachen, Vice President of Tencent Cloud, said: "Star Pulse Network is born for big models. The high-performance network services it provides with large bandwidth, high utilization, and zero packet loss will help break through the bottleneck of computing power, further release the potential of AI, comprehensively improve the training efficiency of enterprise large models, and accelerate the iterative upgrade and application of large model technology on the cloud. ”
The popularity of AIGC has brought about a surge in the number of parameters of AI large models from hundreds of millions to trillions. In order to support large-scale training of massive data, a large number of servers form computing power clusters through high-speed networks, interconnect with each other, and jointly complete training tasks.
Bill Gates: AI can teach kids to read in 18 months, not years
Artificial intelligence has made waves throughout the education sector, and Microsoft co-founder Bill Gates told the media that artificial intelligence chatbots can teach children to read in 18 months, not years. The younger generation is surrounded by digital tools, and experts say it's only a matter of time before classrooms are immersed in AI.
In the future, chatbots, especially supported by large language models of programming, can help students move from elementary education to certification programs, self-directed through a wealth of materials, and tailor their education to specific learning styles.
Through chatbots and AI-powered programs that continuously learn from the information provided to them, teaching can not only be personalized to the needs of students, but this information can provide teachers with more data about students and generate materials instantly. Gradually accept the environment without teachers, accept machine learning.
[This article is from 丨AI Era Network丨WeChat subscription account with the same name, welcome to search and follow]