laitimes

yai | USSR 20000, 2000s 2.0.0moemo.

author:Cutting-edge insights into AI

Roblox releases AI live chat translator

Recently, an AI-based real-time chat translation feature was released, covering 16 languages to facilitate global communication.

According to reports, Roblox has created a translation-specific large language model (LLM), which is trained using "public data and internal data".

It places the LLM in a hybrid expert architecture (MoE), where multiple parallel translation applications are built into a model, each of which handles a language.

Officially, it is mentioned that this method can be used to avoid building a separate translation model for each language while ensuring translation efficiency, thereby saving construction costs and server resources.

yai | USSR 20000, 2000s 2.0.0moemo.

Qingdao Artificial Intelligence Top 10 Events Announced

Recently, the "Top Ten Artificial Intelligence Events in Qingdao in 2023" was announced, covering policies, parks, projects, talents, etc.

1. Qingdao Artificial Intelligence Industrial Park was unveiled

2. A series of policies related to the artificial intelligence industry have been introduced

3. Increase the amount of preservation, and continue to optimize the industrial ecology

4. A series of reports on Qingdao's artificial intelligence industry appeared in the central media

5. The industry model "Made in Qingdao" was unveiled

6. Initiate the establishment of the National Artificial Intelligence Innovation and Application Pilot Zone Industrial Development Community

7. Hosted the finals of the 3rd Shandong Artificial Intelligence Innovation and Entrepreneurship Competition

8. Jiqing jointly held the 5th "100 Enterprises and 100 Scenes" matchmaking fair

9. Normalize the launch of a series of special activities in the artificial intelligence industry

10. Continue to promote the construction of expert teams and industry standards

Microsoft and Semafor launch AI-generated news

According to a Reuters report on February 5, Microsoft plans to cooperate with the American media Semafor to try to write breaking news through AI. Microsoft will sponsor Signals, a new breaking news feed section for Semafor. Semafor's reporters and editors will use Microsoft's AI technology to cover breaking news.

Tiangong 2.0MoE large language model was released

Today, Kunlun Wanwei officially released the new version of MoE large language model Tiangong 2.0 and the new version of Tiangong AI intelligent assistant APP, which is the first 100-billion-level parameter large language model AI application in China equipped with MoE (expert hybrid model) architecture and free for all C-end users. From now on, users can download the Tiangong AI intelligent assistant APP in each mobile application market for experience.

Tiangong 2.0 is the largest version upgrade of Kunlun Wanwei since the release of the double 100 billion-level large language model Tiangong in April last year. This update comprehensively upgrades the answer quality and response speed of AI search, dialogue, reading, and creation, and is equipped with powerful multi-modal capabilities, supporting multi-modal applications such as graphic dialogue, textual and textual diagrams, supporting ultra-long context windows (more than 150,000 Chinese characters) up to 100K, and adding a number of new gameplay methods such as AI painting, data analysis, AI companion, AI fortune telling, and hot meme encyclopedia.

yai | USSR 20000, 2000s 2.0.0moemo.

Zhipu AI launched a creator incentive program

On February 5, Zhipu AI launched a creator incentive plan for Zhipu Qingyan's personalized GLM agent, which comprehensively empowers large model developers from the perspectives of technology, products, security, computing power and investment, so as to better support developers' technical and business practices. The creator incentive program mainly includes GLM API Tokens support, Zhipu Qingyan official recommendation, product/technical/security team support, publicity support, strategy and investment support.

Shenzhen announced the opening of the Shenzhen Intelligent Computing Center

The center lights up operations. The supply and dispatch scale of the Shenzhen Open Intelligent Computing Center and the Shenzhen Smart City Computing Power Coordination and Scheduling Platform will exceed 30,000P, building a high-quality development pattern of computing infrastructure with "advanced technology, reasonable layout, matching supply and demand, safe and reliable, green and low-carbon", and helping to empower the high-quality development of Shenzhen's artificial intelligence industry.

VeriSilicon plans to raise 1.8 billion yuan for AIGC

On February 3, VeriSilicon, a Shanghai semiconductor IP company, issued an announcement on the issuance of A-share prospectus to specific targets in 2023, planning to raise no more than 1.808 billion yuan for the R&D projects of chiplet solution platforms in the field of AIGC and smart mobility, as well as the R&D and industrialization projects of a new generation of IP for AIGC, graphics processing and other scenarios.

yai | USSR 20000, 2000s 2.0.0moemo.

Lenovo, Microsoft, and others signed an agreement on AI ethics

Recently, eight global technology companies – Lenovo Group, Microsoft, LG AI R&D, INNIT, SAFTSE and Telefónica – signed an agreement in Slovenia committing to UNESCO's recommendations on the ethics of AI.

The agreement was signed at the second UNESCO Global Forum on Artificial Intelligence in Kranj, Slovenia. It requires companies to play a full role in protecting human rights in the design, development, purchase, sale and use of AI, and this is the first time that companies have engaged with the United Nations in this area.

It states that businesses must conduct due diligence to identify the adverse effects of AI in order to meet safety standards and take timely steps to prevent, mitigate or remediate these impacts in accordance with domestic legislation. The agreement also states that pre-testing (testing new AI systems before they are put on the market) is crucial.

But given the rapid development of AI systems already on the market, it also requires the development of post-event (post-deployment) risk assessment and mitigation practices.

AI startup Meshy launched the Meshy-2 Wensheng 3D image model

On February 6, AI startup Meshy announced the release of its second-generation product, Meshy-2. Compared to the first generation, Meshy-2 is capable of producing better meshes and rich geometric details. In addition, Meshy-2 also supports four text-to-3D graphics creation styles: realistic, cartoon, low-poly, and voxel. In terms of efficiency, Meshy-2 can produce a graphical preview in just 25 seconds and detailed results in 5 minutes.

The 1.5 version of the Tongyi Qianwen model is open source

According to GitHub, version 1.5 of the Tongyi Qianwen model (Qwen) was launched on February 6, including six model sizes: 0.5B, 1.8B, 4B, 7B, 14B and 72B, of which the strongest version surpasses GPT-3.5 in performance, and related technologies have been launched on the official website of Tongyi Qianwen and the Tongyi Qianwen App.

宜家AI家居助手上线GPT Store

On February 6, home furnishing giant IKEA announced the launch of a new IKEA AI assistant on its official website. This is an AI-powered tool that integrates home design, inspiration, and shopping, and has been launched on the OpenAI GPT Store. Users can use GPT to clarify specific home design needs and get AI-tailored recommendations.

复旦推出AI Agent视障助手

FudanNLP, a natural language processing laboratory at Fudan University, announced on its WeChat official account platform on February 6 that they have launched the "Hear the World" Agent based on the multi-modal large model MouSi to help the visually impaired group re-perceive the world. At present, the app has a street walking mode, a free question and answer mode, and a finding mode.

Nature:AI破解2000年前希腊语卷轴

On February 5, participants in the Vesuvius Challenge used AI to uncover Greek text from scrolls dating back 2,000 years through training learning algorithms, and their findings were published in Nature. Currently, researchers have published hundreds of words in 15 lines of text through TimeSformer's model, which equates to about 5% of the entire scroll. Researchers say this could have a revolutionary impact on our understanding of the ancient world.

For more AI information, please pay attention to the official account "AI Intelligent Sharing Frontier"

Reply to "Reports" to get the latest industry reports

Read on