"Geek Headlines" - the news circle for technicians!
Good morning, readers and friends of CSDN, "Geek Headlines" is here, let's see what are the important news worthy of our technical people's attention today.
Organize | Meng Yidan
Listing | CSDN (ID: CSDNnews)
A quick look at news points in one minute!
-
Li Yanhong: The reasoning speed of the Wenxin large model version 3.5 has increased by 17 times
-
Tencent Cloud fully disclosed the self-developed Velar high-performance computing network for the first time
-
Station B will use the playing time instead of the playing times
-
Li Yanhong, Zhou Hongyi, and Zhang Yong gathered in the former residence of Confucius, Zhou Hongyi: Three old friends agree on AI security
-
Zhu Xiaohu and Fu Sheng's circle of friends "mutually hate each other": Should entrepreneurs compete in large models?
-
Databricks acquires MosaicML for $1.3 billion
-
The growth rate of ChatGPT visits dropped sharply
-
Japan plans to ban candidates from using ChatGPT
-
Chip giant Intel announces major restructuring plan
-
IBM Announces $4.6 Billion Acquisition of Apptio
-
Linux 6.4 released
-
Go 1.21 Release First Release Candidate (RC)
domestic news
Li Yanhong: The reasoning speed of the Wenxin large model version 3.5 has increased by 17 times
When Robin Li attended the Nishan Dialogue on Digital Civilization at the World Internet Conference on June 26, he believed that the key point of the new international competition strategy is not how many large models a country has, but how many native AI applications there are on the large models, and how big these applications are. Increased production efficiency to a certain extent. At the same time, Li Yanhong revealed that the Baidu Wenxin model has been iterated to version 3.5. Compared with version 3.0, the training speed has been increased by 2 times, the reasoning speed has been increased by 17 times, and the model effect has been improved by more than 50%. The data quality, generation effect and content security have all been significantly improved. (36 krypton)
Tencent Cloud fully disclosed the self-developed Velar high-performance computing network for the first time
Tencent Cloud fully disclosed the self-developed Velar high-performance computing network for the first time. Xingmai network has the industry's highest 3.2T communication bandwidth, which can increase GPU utilization by 40%, save model training costs by 30%~60%, and bring a 10-fold improvement in communication performance for large AI models. Based on Tencent Cloud's new-generation computing power cluster HCC, it can support a super-large computing scale of 100,000 cards.
Station B will use the playing time instead of the playing times
In the live broadcast of the 14th anniversary of Station B, Chen Rui, chairman and CEO of Station B, delivered a speech entitled "Nice to meet you". Chen Rui said that the number of minutes played can better reflect the quality of the video than the number of times played, but the technical complexity and cost of counting the number of minutes played is higher. Fortunately, with the maturity of technology, Bilibili has the ability to count the number of minutes played, and replace the number of times played by Bilibili as explicit data, so that users have better reference. "It is planned that in the next few weeks, product updates will be completed." (Caijing Technology)
Li Yanhong, Zhou Hongyi, and Zhang Yong gathered in the former residence of Confucius, Zhou Hongyi: Three old friends agree on AI security
Zhou Hongyi said, “In the hometown of Confucius, we discussed how to build a safe, reliable, controllable and easy-to-use artificial intelligence. The three old friends agreed.” It is to help employees improve their capabilities and efficiency." (Sina Technology)
Zhu Xiaohu and Fu Sheng's circle of friends "mutually hate each other": Should entrepreneurs compete in large models?
Fu Sheng, CEO of Cheetah Mobile, and Zhu Xiaohu, managing director of GSR Venture Capital, started a mutual confrontation in the circle of friends. The reason was that Fu Sheng reposted a related article about Zhu Xiaohu’s denial of large-scale entrepreneurship, with the text “Half of the start-ups in Silicon Valley revolve around ChatGPT Let’s start, our investors can still be so ignorant and fearless.” Then Zhu Xiaohu himself left the comment area, and the core debate was “should Chinese entrepreneurs compete for big models, and what value can they create”. (Titanium Media)
international news
Databricks acquires MosaicML for $1.3 billion
Big data giant Databricks announced that it has signed a definitive agreement to acquire San Francisco-based artificial intelligence startup MosaicML for $1.3 billion. Following the acquisition, MosaicML will become part of the Databricks Lakehouse platform, bringing MosaicML's entire team and technology under the Databricks umbrella, providing enterprises with a unified platform to manage data assets and the ability to build, own and secure with their own proprietary data Own generative AI models.
It is worth noting that MosaicML is a very young generative AI company. It was founded in San Francisco in 2021. It has only publicly disclosed one round of financing and has only 62 employees. In the last round of financing, its valuation was 220 million US dollars, that is to say, the valuation of the acquisition of MosaicML directly jumped 6 times. (Silicon Starman)
The growth rate of ChatGPT visits dropped sharply
According to news on June 26, ChatGPT, which has set off an AI boom in the world, seems to have entered a bottleneck period. According to data from the website data analysis tool SimilarWeb, the growth rate of ChatGPT's visits in the early stage was astonishing. The month-on-month growth rate was 131.6% in January, 62.5% in February, and 55.8% in March. It slowed down significantly in April, with a month-on-month growth rate of 12.6%, and by May, this figure had become 2.8%. The month-on-month growth rate in June is likely to be negative. (one fortune)
Japan plans to ban candidates from using ChatGPT
According to Japanese media reports on June 22, people familiar with the matter revealed that the Ministry of Education, Culture, Sports, Science and Technology plans to implement new guidelines instructing elementary, middle and high schools to prohibit students from using generative artificial intelligence such as ChatGPT in exams. (AI) software. MEXT did not completely rule out the use of such tools, though, and the draft guidelines list examples of their effective use, such as facilitating class discussions, correcting grammar in English conversations or learning advanced programming skills. MEXT will issue the guidelines as early as July, after listening to stakeholders and making necessary revisions, the source reportedly said. (Reference News)
Chip giant Intel announces major restructuring plan
Recently, Intel announced that the foundry business will become an independent department. In an online analyst meeting, Intel stated that it is adjusting its corporate structure. It plans to operate the wafer foundry business (IFS) independently in the first quarter of next year, and list the profit and loss (P&L) separately in the financial report. This is called a new "internal foundry model" by Intel. Intel CEO Pat Kissinger proposed to promote the IDM2.0 model in 2021 and decided to enter the foundry industry. Wait and compete.
Intel also said that the internal foundry model can promote Intel's goal of saving US$3 billion in costs by 2023 and US$8 billion to US$10 billion in costs by 2025. At the same time, it will also help Intel achieve its long-term growth target of 60% non-GAAP gross profit margin and 40% operating profit margin. (21st Century Business Herald)
IBM Announces $4.6 Billion Acquisition of Apptio
IBM announced Monday that it will acquire software company Apptio from US private equity firm Vista Equity Partners for $4.6 billion. The acquisition of Apptio, which provides financial and operational IT management and optimization tools, is a continuation of IBM's investments in artificial intelligence and IT automation software, the company said. Apptio customers include Amazon, Microsoft, and Google, among others.
Programmer Zone
Linux 6.4 released
Linus Torvalds announced Linux 6.4 on the kernel mailing list, and the Linux 6.5 merge window opened, but in fact a large number of Linux 6.5 merge requests were submitted last week. The Linux author, who previously criticized the kernel developers for delaying merge requests until the last week, now expresses his gratitude to these proactive developers. The main new features of Linux 6.4 include: BPF general iterators, disable SELinux runtime disable, remove SLOB memory allocator, Intel CPU linear address mask support, user trace event support, continue to improve Rust support, per-VMA locks, etc., More can be found at https://kernelnewbies.org/Linux_6.43
Go 1.21 Release First Release Candidate (RC)
The Go official blog officially released the first candidate version of 1.21. Now developers can download and try it out, and give feedback so that the team can release the final version in August. It's worth mentioning that the first RC for Go 1.21 is called go1.21rc2 because bugs were found and fixed after tagging go1.21rc1. This version mainly focuses on enhancements and improvements in tools, languages, standard library plug-ins, and performance improvements. Among them, the Profile Guided Optimization (PGO) function is now officially launched. If a file named default.pgo exists in the main package directory, the go command will use it to enable PGO builds. The go tool supports forward and backward language compatibility. Added built-in functions: min, max and clear, etc. For details, please refer to the release notes: https://go.dev/blog/go1.21rc