Abstract: It has been half a year since the release of Midjourney v5. When will the next major version v6 be released? The answer is: later this year. Midjourney CEO David Holz said at a recent event that the leap from Midjourney v5 to v6 will be greater than the leap from v4 to v5. However, Holz did not give an exact release date, but said it will be released soon, but it will definitely be this year.
Image source: Generated by Unbounded AI
Hot info
Two High Schools and One Ministry of Education: Using generative AI technologies such as "deep synthesis" to commit cyber violence crimes shall be severely punished in accordance with the law.
Babbitt News, on September 25, the Supreme People's Court, the Supreme People's Procuratorate, and the Ministry of Public Security jointly issued the "Guiding Opinions on Punishing Internet Violence Violations and Crimes in accordance with the Law." The "Opinions" require that those who commit cyber violence crimes under any of the following five circumstances shall be severely punished in accordance with the law: those who commit them against minors or disabled people; those who organize "trolls", "thugs" or other personnel to commit them; those who fabricate "Sex-related" topics infringe on the human dignity of others; use "deep synthesis" and other generative artificial intelligence technologies to publish illegal information; initiated and organized by network service providers.
Microsoft joins forces with Johns Hopkins University to launch ALMA, a large model focused on translation work
According to Pinwan’s report on September 25, the Arxiv page shows that researchers from Johns Hopkins University recently teamed up with Microsoft to launch a new large language model ALMA, focusing on translation work. The full name of ALMA is Advanced Language Model-based trAnslator, which is developed based on LLaMA-2. The model introduces a new fine-tuning method that eliminates the need for large amounts of parallel data that traditional translation models often rely on through initial fine-tuning on monolingual data and then subsequent fine-tuning on a small set of high-quality parallel data. Experimental results show that ALMA performs better than traditional models in translation work.
Baichuan Intelligent releases Baichuan2-53B large model to start commercialization for the B-end market
According to the Science and Technology Innovation Board Daily report on September 25, Baichuan Intelligent released the Baichuan2-53B closed-source large model, which improves mathematical and logical reasoning capabilities and reduces model illusions through high-quality data systems and search enhancements. This time Baichuan Intelligence also opened the API interface and officially entered the ToB field, which will start the commercialization process. It is understood that Baichuan2-53B is the sixth large model released by Baichuan Intelligence. As previously reported, Baichuan Intelligent released the Baichuan 2 large model.
Du Xiaoman’s “Xuanyuan 70B” large financial model is open source
According to a report by Webmaster Home on September 25, Du Xiaoman launched a new large-scale model for the financial industry "Xuanyuan 70B". "Xuanyuan 70B" is an important large-scale financial dialogue model project designed to provide excellent natural language processing capabilities. This project is built on the Llama2-70B model. After Chinese enhancement and a large number of pre-training in the general and financial fields, a series of powerful financial large models have been formed.
"Xuanyuan 70 B" is an open source large model with a parameter magnitude of 70B and a context length of more than 8k. It incorporates a large amount of professional financial corpus in the pre-training stage, aiming to improve general capabilities while significantly enhancing financial professional performance. This model has been fully opened in the open source community, including model weights and training corpus, providing an efficient Chinese base model for the financial industry. Users can also perform customized fine-tuning as needed.
Alibaba Cloud Zhou Jingren: The new open source large model Qwen-14B provides more powerful cognitive and other capabilities
According to 36 Krypton’s report on September 25, Alibaba Cloud Intelligence Chief Technology Officer Zhou Jingren announced at the Tongyi Qianwen Open Source Conference that a new open source large model with stronger performance, Qwen-14B, will be based on 14 billion parameters and 3 trillion The open source large model trained by token provides more powerful reasoning, cognition, planning and memory capabilities. In addition, compared with the previous open source version, the new QWEN-7B (v1.1) model has generally improved capabilities, with core indicators increased by up to 22.5%.
Writer, an AI startup that raised $100 million, open sourced its large language model Palmyra
Babbitt News, according to the public account "AIGC Open Community", the artificial intelligence startup Writer has open sourced its large language model Palmyra on huggingface. There are 8 models, namely small, base, 20b-chat, Instruct-20b, med -20b, etc., are commercially available and support data fine-tuning.
According to reports, Palmyra’s technical highlights include powerful small functions, which are very helpful for small and medium-sized enterprises and individual developers without computing resources; it has received training in business writing and marketing data, and is mainly for enterprise users; enterprise-level data security, built-in multiple Security Fence. In addition to generating text, it can also extract content summaries of videos, PDFs, and audios; it supports data fine-tuning, and enterprises can create their own "ChatGPT" assistant, etc. As previously reported, AI startup Writer completed US$100 million in financing and was valued at more than US$500 million.
Investment and Financing
Amazon to invest $4 billion in AI startup Anthropic
Amazon said on Monday it would invest $4 billion in cash in artificial intelligence startup Anthropic to compete with growing artificial intelligence cloud competitors, Devdiscourse reported. As part of the deal, Amazon employees and cloud customers will have early access to Anthropic's technology and integrate it into their own businesses.
Anthropic has also pledged to rely heavily on Amazon's cloud services, including using large quantities of proprietary chips purchased from Amazon to train its future artificial intelligence models. The CEOs of Amazon's cloud unit and Anthropic said in a joint interview that the immediate investment would be $1.25 billion, with both parties having the right to trigger another $2.75 billion in financing from Amazon. Additionally, they declined to say how much stake Amazon will now own in Anthropic, or the startup's latest valuation.
It is reported that Anthropic was founded in 2021 by former OpenAI researchers and is headquartered in San Francisco, USA. The company developed the chatbot Claude and received investment from technology giants such as Google and Salesforce. It completed US$450 million in Series C financing in May. It was previously reported that Anthropic, an American AI company backed by Google, received a US$100 million investment from South Korea's SK Telecom.
TTC, an AI-native international human resources service provider, completed tens of millions of yuan in Pre-A round of financing
According to a report by the Science and Technology Innovation Board Daily on September 25, TTC, an AI-native international human resources service provider, recently completed tens of millions of yuan in Pre-A round financing. The leading investor in this round of financing is Sinovation Ventures.
Recommended Reading of the Metaverse
"Only 1% of parameters are needed, the effect surpasses ControlNet, the new AI painting control master is coming"
"New AI painting detail control master" ControlNet-XS is here! The important thing is that the parameters are only 1% of the original ControlNet. ControlNet-XS was developed by the Computer Vision Laboratory of Heidelberg University. Currently, relevant papers and pre-training models have not been released. But the researchers said that ControlNet-XSFID scored significantly better than ControlNet. And the code that controls Stable Diffusion-XL and Stable Diffusion 2.1 will be open source in the near future.
https://www.8btc.com/article/6833834
"China and the United States "encircle and suppress" Nvidia, it is difficult to dominate the trillion-dollar AI chip market"
As Nvidia takes the lead in occupying the AI large model computing power chip market, it has a market value exceeding one trillion. At the same time, Intel, AMD and Chinese GPU chip companies are also secretly competing, hoping to get a share of the AI computing power chip market.
https://www.8btc.com/article/6833824
"In response to the challenge of OpenAI, Midjourney will release the V6 version upgrade this year"
It has been half a year since the release of Midjourney v5. When will the next major version v6 be released? The answer is: later this year. Midjourney CEO David Holz said at a recent event that the leap from Midjourney v5 to v6 will be greater than the leap from v4 to v5. However, Holz did not give an exact release date, but said it will be released soon, but it will definitely be this year. As for what new things Midjourney v6 will bring, Holz promised to bring better text understanding, that is, image generation will be closer to the description of the prompt, and better recording of the details in the prompt wording, and may also bring text rendering. Function.
https://www.8btc.com/article/6833878
Babbitt Park is open for cooperation!
Chinese Twitter: https://twitter.com/8BTC_OFFICIAL
English Twitter: https://twitter.com/btcinchina
Discord community: https://discord.gg/defidao
Telegram channel: https://t.me/Mute_8btc
Telegram community: https://t.me/news_8btc
