According to The Verge, ByteDance is secretly using OpenAI's technology to develop its own large language model (LLM). In the field of artificial intelligence, this practice is often considered inappropriate. This is also a direct violation of OpenAI's Terms of Service. When ByteDance used GPT to train their AI model, OpenAI suspended their account. However, most of ByteDance's GPT usage is done through the Microsoft Azure platform, not directly through OpenAI. It's unclear whether Microsoft will follow OpenAI's lead and suspend access to ByteDance.
According to the list published by the U.S. Trademark and Patent Office (USPTO) on the 14th of this month, Samsung has obtained a new trademark called "Flex Magic", which is implied to be used in the next generation of XR headsets. In the trademark description, Samsung writes that the trademark applies to products such as 3D glasses, virtual reality headsets, virtual reality goggles, and smart glasses. It is reported that Samsung has already applied for the same trademark with the European Patent Office in November.
Although there is no direct connection between the trademark application and the actual application, it at least indicates that Samsung is actively promoting XR headsets. According to South Korean industry sources, Samsung plans to release an XR headset in the second half of next year, codenamed "Infinite", and plans to go on sale by the end of next year.
On December 15, NetEase Youdao announced the 2Version 0. According to reports, the new version has undergone four major ability innovation upgrades: adding a new oral difficulty classification; Richer avatars; More diverse dialogue scenarios and more personalized dialogue evaluation reports.
Perplexity can now reportedly generate based on users' searches and results. The CEO said they were about to launch an image generation service. Some users said that they had tried it out and found that this feature was indeed online. In this way, users can sort out a generated ** as a header image after searching for the content, and then publish it directly.
This feature should be using the DALL-E3 technology, and the generated ** will be marked with AI in the lower right corner. After the search is completed, users can click the "Generate Image" button in the lower right corner, and then select the style, including painting, illustration and diagram.
Alibaba reportedly released a ** in November, announcing that it would open-source the i2 VGEN-XL image generation** model. Now, they have finally released concrete ** and models. This model can generate a ** presentation without large character movements.
The i2 VGEN-XL model is divided into two phases. The first is the foundation phase, which guarantees coherent semantics through the use of two hierarchical encoders and preserves the content of the input image. This is followed by the optimization phase, which enhances the details by incorporating additional short text and increases the resolution to 1280 x720.
A few days ago, the "2023 Talent Migration Report" released by Maimai Gaopin shows that in 2023, the new economy will continue to be saturated with talents, and the talent supply and demand ratio will increase from 032 rises to 2 in 202304, an average of 2 people compete for 1 position. However, AI-related positions are in short supply, and among the top 20 high-paying positions, there are 10 AI positions represented by ChatGPT researchers, algorithms, and deep learning Xi. ChatGPT researchers earn an average monthly salary of 670,000 yuan tops the list of high salaries.
Tang Xiaoou, chairman of SenseTime and an artificial intelligence scientist, died in his sleep on December 15 at the age of 55. According to public information, Tang Xiaoou was born in Anshan, Liaoning Province in 1968, and is a professor in the Department of Information Engineering of the University of Hong Kong Chinese and an outstanding scholar in the School of Engineering. Xiaoou Tang is mainly engaged in research in computer vision-related fields, including multi-vision, computer vision, pattern recognition and processing.
On December 16, the National Language Resources Monitoring and Research Center released the "Top Ten New Words in China in 2023". The top 10 new words released this time are: Generative Artificial Intelligence, Global Civilization Initiative, Village Super, New Quality Productivity, National Ecology Day, Consumption Boost Year, Special Forces Tourism, Conspicuous Package, Hundred Model War, and Mozi Sky Survey.
According to interpretation, generative artificial intelligence is a new type of artificial intelligence that generates new original content by learning Xi large-scale datasets, and it is a technology that generates text, sound, and other content based on algorithms, models, and rules.
The 100-model war refers to the trend of various "large-scale deep Xi models" competing for development in the application field. Since 2023, more than 100 large models of various types have been released in China, and these large models and their products are mainly divided into three categories: the first category is general large models; The second type is the industry model; The third type is the application service model based on the general model or industry model.
Tianyancha shows that recently, Weilai Automobile Technology (Anhui)** has undergone industrial and commercial changes, and its business scope has added electric vehicle charging infrastructure operation, new energy vehicle sales, artificial intelligence basic software development, integrated circuit chip design and services, etc. The company was established in August 2020, the legal representative is Qin Lihong, the registered capital is 6 billion yuan, and it is wholly owned by NIO Holdings***.
On December 15, Hotgen Biotech announced that the company recently established the "X-Gen AI New Drug Discovery and Design Research Center", which is expected to promote the application of artificial intelligence technology in the field of drug research and development and accelerate the process of new drug discovery and design.
Google's DeepMind recently announced a model training method called "funsearch", which claims to be able to calculate a series of "complex problems involving mathematics and computer science", including "ceiling-level problems" and "boxing problems".
It is reported that the FunSearch model training method mainly introduces an "evaluator" system for AI models, the AI model outputs a series of "creative problem-solving methods", and the "evaluator" is responsible for judging the solution methods output by the model, and after repeated iterations, AI models with stronger mathematical ability can be trained.
Google DeepMind used the Palm 2 model for testing, and the researchers set up a dedicated "pool", using the form to input a series of questions to the model, and set up an evaluator process, after which the model will automatically select questions from the pool in each iteration to generate a "creative new solution", which will be evaluated by the evaluator, where the "best solution" will be re-added to the pool to start another iteration.
On December 16, at the Geek Park Innovation Conference 2024, Robin Li, Founder, Chairman and CEO, said, "On top of the basic model, there must be thousands or even millions of AI-native applications for the value of this large model to be reflected." In the past year, the main excitement of the society and the public is still on the basic model, and I have not moved to the native application of AI, and I am more or less anxious, so the last few public speeches, including this kind of speech within the company, are constantly emphasizing that we must roll up the native application of AI, and we must make this thing to make it valuable. ”
According to the ARXIV page, Tencent recently joined hands with Xi'an Jiaotong University and the University of Hong Kong to jointly publish**, introducing a multi-modal large model VL-GPT. **Indicates that VL-GPT is a transformer model capable of perceiving and generating both visual and verbal data. A unified pre-training approach for image and text patterns is achieved by employing direct automatic regression goals, allowing the model to process images and text as seamlessly as a language model processes text.
Studies have shown that VL-GPT has demonstrated excellent performance in a variety of visual and language understanding and generation tasks, including image captioning, visual problem solving, text-to-image generation, and more.
Spotify recently confirmed that it is testing a prompt-based AI playlist feature that allows users to create playlists using AI technology and prompts. **Shows the process of creating a playlist using ChatGPT in the Spotify app via the "Your Library" option, with AI responding to the user's prompts and generating a playlist. Spotify confirmed the test, but did not disclose technical details, how it works, or promise when it will be officially launched.
Li Auto announced that its self-developed Mind GPT Chinese large model won the first place in the C-Eval Chinese Large Model Comprehensive Evaluation List and CMMLU Chinese Large Model Comprehensive Evaluation Benchmark.
It is understood that Li Auto's self-developed multi-modal cognitive model Mind GPT is built around in-vehicle scenarios and has the ability to understand, generate, memorize and reason. Based on the key scenarios of ideal students, Mind GPT has tailored more than 1,000 exclusive abilities covering 111 fields.