AI Weekly Report #14 Newer models, stronger models
This week’s cover: @ victor1112
Prompt word: beautiful girl of lahore
Alibaba Tongyi Qianwen 2.0 is released, with model parameters reaching 100 billion levels
According to news on October 31, at the 2023 Yunqi Conference, Alibaba Cloud Chief Technology Officer Zhou Jingren announced the release of a comprehensive upgrade of Alibaba’s large model family, and the official upgrade and release of Tongyi Qianwen 2.0.
ChatGPT tests new feature: automatically selects relevant tools to use based on context
OpenAI is currently launching a new beta feature for ChatGPT Plus members. According to user feedback, this update includes the ability to upload and process files, as well as multi-modal support. This means that users don't need to select a mode like Browse with Bing from GPT-4's drop-down menu, but instead automatically guess what the user wants to do based on context.
These new features allow ChatGPT Plus members to also have some office functions provided by the ChatGPT Enterprise plan. It is reported that after the file is passed to ChatGPT, it takes a few seconds to digest the file, and then it can process the file, such as summarizing the data, answering questions or generating data visualizations according to prompts.
Gen-2 release update brings major improvements to fidelity and consistency
On November 3, generative AI video startup Runway announced the release of the Gen-2 update, which brings significant improvements to the fidelity and consistency of video results. According to @TomLikesRobots, an AI artist on X, the resolution of Gen-2's video generation from still images has been upgraded from 1792×1024 to 2816×1536.
Midjourney updates launch Style Tuner to create custom styles
On November 3, Midjourney launched a new feature called "Style Tuner" that allows users to create their own styles and apply them to new image generation.
Style Tuner affects various aspects of the generated image, such as color and font details. Users can adjust to specific commands or settings, aspect ratios, and even image and text cues in raw mode. Users can share their adjusted styles with others.
Sophon Engine releases multi-modal large model "Yuan Cheng Xiang Chatimg 3.0"
At the CNCC 2023 "Super Intelligence Fusion AI Large Model Application Implementation Development Forum" held on October 28, Sophon Engine released "Yuancheng Xiang Chatimg3.0" and demonstrated the multi-modal general generation model "Yuancheng Xiang Chatimg3.0. 0"'s latest progress and implementation exploration.
Yuanchengxiang Chatimg3.0 is a large multi-modal model with ultra-fine recognition and less hallucinations. It also supports multi-image understanding, object positioning, OCR and other functions. Chatimg3.0 equips hardware devices with brains, enabling more natural and smooth human-machine communication, laying a solid foundation for AI multi-modal large model-empowered industrial applications.
Baichuan Intelligent launches Baichuan2-192K, the world's longest context window model
On October 30, Baichuan Intelligent released the Baichuan2-192K large model, with a context window length of up to 192K, which is currently the longest context window in the world. Baichuan2-192K can process about 350,000 Chinese characters at a time, which is 4.4 times that of Claude2, the best large model that currently supports long context windows (supports 100K context windows, measured about 80,000 characters), and is even more powerful than GPT-4 (supports 32K context windows, measured 14 times that of about 25,000 words).
Baichuan2-192K not only surpasses Claude2 in context window length, but also leads Claude2 in terms of long window text generation quality, long context understanding, long text Q&A, and summary. At present, Baichuan Intelligent has launched the API internal test of Baichuan2-192K, which is open to core partners in the legal, media, financial and other industries.
Vivo releases self-developed general AI large model Blue Heart model
According to news on November 1, at the 2023 vivo Developer Conference, vivo officially released its self-developed general AI large model matrix-Blue Heart Large Model. According to reports, the Blue Heart model will include a total of 5 models with three parameter levels: billion, tens of billions, and hundreds of billions, fully covering core scenarios, and leading the industry in model capabilities. Vivo believes that a truly good self-developed large model has five points - large and comprehensive, strong algorithm, truly safe, self-evolving, and widely sourced.
Among them, the first of the two 1 billion-level models is the 7 billion-level Blue Heart Large Model 7B. This is a device-cloud dual-use model built for mobile phones. Vivo claims that it performs very well in scenarios such as language understanding and text creation. . The other is the 1 billion-level blue heart large model 1B, which is a professional text large model built for client-side scenarios. It has localized text summary, abstract and other capabilities, and the processing process is accurate and safe.
Wenxinyiyan launches membership: 49.9 yuan for a continuous monthly subscription, which can unlock Wenxin Big Model 4.0
On November 1st, Baidu Big Model Wenxin Yiyan launched a membership model, which can unlock Wenxin Big Model 4.0 after activation.
According to reports, the single-month purchase price of Wenxinyiyan membership is 59.9 yuan, and the continuous monthly subscription price is 49.9 yuan (automatically renewed at 49.9 yuan/month the next month, and can be canceled at any time). In addition, Baidu also launched a joint membership of Wenxin Yiyan + Wenxin Yige (Silver), priced at 99 yuan/month.
Netizen Javi Lopez used GPT-4, DALL·E and Midjourney to successfully reproduce "Angry Birds" and create this little game called "Angry Pumpkin".