Dynamics of AI and Blockchain

The Shanghai Municipal Government recently issued a document proposing to accelerate the construction of high-performance computing and high-quality data infrastructure to support artificial intelligence large models and blockchain innovative applications. The city plans to build a diverse supply, cloud-edge collaboration, on-demand scheduling, and efficient green urban high-performance computing network system. The goal is to establish intelligent computing resources that can support trillion-level parameter large model training, a high-quality corpus, and specialized datasets, and to initially establish urban blockchain infrastructure centered around the Pujiang Data Chain.

The artificial intelligence company OpenAI is in discussions with potential investors, planning to sell existing employee-held shares at a valuation of $86 billion. It is understood that the specific allocation plan has not yet been finalized, and related terms may still be subject to adjustments.

The image generation company Midjourney has partnered with the Japanese game company Sizigi Studios to launch a mobile application called NijiJourney, which primarily provides AI-generated images in anime style for the Japanese market. The application operates on a subscription model, allowing users to choose between a one-time annual payment of $96 or a monthly payment of $10. Existing Midjourney users can log in using their Discord credentials without any additional fees.

OpenAI CEO Sam Altman stated at a recent tech event that while he is very interested in AI-based consumer electronics, he is currently unsure about the specific form these devices will take. He believes that future AI devices may not be as ubiquitous as smartphones.

Recently, Jiadu Technology Group successfully won the bid for the Ministry of Industry and Information Technology's "2023 Industrial Technology Infrastructure Public Service Platform - Public Service Platform Project for Engineering Technology and Applications of Artificial Intelligence Large Models." This project is the first major special project for large models by the Ministry of Industry and Information Technology and is also the only major special project for public service platforms for large models this year. Jiadu Technology has formed a consortium with several top research institutions, universities, and enterprises, becoming the first candidate.

American AI company Jasper AI announced the launch of a new end-to-end AI-assisted marketing tool aimed at enhancing marketing effectiveness. The new features include content optimization analysis, brand strategy consistency management, and marketing campaign tools that accelerate the review process. These features will begin testing in November and are planned for a full launch in the first quarter of 2024.

The three major music publishers Universal, Concord, and ABKCO have recently sued the AI startup Anthropic for copyright infringement. The lawsuit points out that Anthropic illegally copied and distributed a large number of copyrighted works, including the lyrics of music works owned or controlled by these publishers, while developing and operating its AI models.

ChatGPT officially introduces web search capabilities, allowing real-time access to the internet and providing information source links. This feature is currently available only to ChatGPT Plus and Enterprise subscription users.

Meta AI has released a new artificial intelligence system capable of decoding images from human brain activity. This system combines non-invasive magnetoencephalography (MEG) scanning technology to reconstruct the images perceived and processed by the brain in real-time. However, this experimental system requires pre-training on individual brain waves and currently cannot produce thought images unrelated to model training.

Alluxio has launched a new generation enterprise AI platform that increases the efficiency of GPU devices by 4 times through caching technology. The platform adopts data virtualization and DORA architecture, improving the overall model training process speed by 20 times, and optimizes the data transfer from storage to GPU using hardware acceleration technology.

According to reports, OpenAI has halted the development of a more efficient GPT-4 level model codenamed "Arrakis." The model was originally planned to reduce deployment costs while maintaining capabilities comparable to GPT-4, using a sparse design principle. However, the development team found that the performance did not meet standards after starting training this spring, and after a month of adjustments, they ultimately decided to terminate the project.