The Basic Model Research Center of the Institute of Artificial Intelligence of Tsinghua University was established

Big Model Information:

According to the official news of Tsinghua University, on the afternoon of June 30, the establishment ceremony of the Basic Model Research Center of the Institute of Artificial Intelligence of Tsinghua University was held in the conference room of the Institute of C.

Wang Xiqin, president of Tsinghua University, said that basic models are crucial to the development of general artificial intelligence, and the establishment of a research center for the world's scientific and technological frontiers aims to promote organized scientific research and discipline development in related fields. It is hoped that the research center can strengthen overall planning and actively explore innovative governance models. It is necessary to make good use of Tsinghua's comprehensive disciplinary advantages, gather the resources of faculties, departments and departments, and form a joint force in the same direction to jointly promote the innovation and application of artificial intelligence basic models, and make due contributions to the independent training of key core technologies and top-notch innovative talents in the mainland.

Zhang Yaoxue, dean of the Institute of Artificial Intelligence of Tsinghua University, pointed out that the Institute of Artificial Intelligence of Tsinghua University has a good accumulation of preliminary work in the research and application of artificial intelligence basic models, and has built a number of influential basic models and made a good start in industrialization. The establishment of the research center will focus on the research and application of artificial intelligence basic models, and contribute to the construction of artificial intelligence-related disciplines, the improvement of scientific research level and the transformation of scientific research achievements in schools.

Tencent Cloud officially released the vector database

Tencent Cloud officially released AI Native vector database Tencent Cloud VectorDB. The database can be widely used in scenarios such as training, inference, and knowledge base supplementation of large models, and is the first vector database in China that provides full-lifecycle AI from the access layer, computing layer, and storage layer.

According to reports, Tencent Cloud Vector Database supports up to 1 billion-level vector retrieval scale, and the latency is controlled in milliseconds, which is 10 times higher than the traditional stand-alone plug-in database retrieval scale, and has a peak capability of million-level queries per second (QPS).

Meta's new product Threads has been launched on the APP Store and is expected to be released on July 6

According to Deadline, Meta's new social media product Threads is now available on the App Store and will be officially released on July 6.

It is reported that Threads is a social media software with the concept of decentralization, users can post opinions in Threads and communicate with other users. Instagram users can log in directly to Threads without having to sign up for a new account.

It is worth mentioning that Musk said last week that he would limit the number of Twitter users read per day, which led many Twitter users to choose to leave. The listing of Threads may attract this part of the user to join.

Urtopia launches Fusion, the first e-bike with ChatGPT

According to insideevs, Urtopia recently launched a smart electric bike Fusion, which will be equipped with ChatGPT for users.

Fusion allows the rider and the bike to talk, and he answers the cyclist's questions and responds accordingly. Fusion can also connect with apps like Apple Health to record ride data. And built-in artificial intelligence can analyze this data and adjust bike settings to improve efficiency.

Urtopia has not yet disclosed the release date and price of the Fusion.

Alibaba's new patent can evaluate and guide the optimization of large models

According to interface news reports, on July 4, Alibaba (China) Co., Ltd. applied for a patent for "data processing method and server based on large model" was announced.

The patent abstract shows that the application obtains the instruction set of human-computer interaction and the annotation information of the instruction in the instruction set, enters the instructions into the large model to realize human-computer interaction, and outputs the response results of the instructions through the large model;

According to the response results and annotation information, determine the evaluation value of the large model in identifying the security risk dimension and the capability dimension of each task, evaluate the performance of the large model accurately, comprehensively and fine-grainedly, and output the evaluation value of the large model to guide the online judgment of the large model or update the optimized version of the large model, so as to improve the accuracy of human-machine dialogue based on the large model and ensure the quality of human-machine dialogue.

Nanyang Technological University and Microsoft unveil DisCo, an AI model

According to the huggingface page, Nanyang Technological University recently announced a video synthesis product DisCo with Microsoft Azure AI.

DisCo can generate a video from an image and make the characters in the picture dance. The paper shows that today's dance video generative models often struggle to solve the gap between synthetic content and real-world dance scenes. DisCo solves this problem better.

Experimental results show that DISCO can generate high-quality human dance images and videos with diverse appearances and flexible movements.

Peking University released ChatLaw, a large model Chinese law

According to the Science and Technology Innovation Board Daily, the Peking University team recently released ChatLaw, a Chinese legal model, to provide inclusive legal services to the public. The model supports file and speech output, as well as legal document writing, legal advice, and legal aid recommendation.

Based on the general large model, the Peking University Shenzhen Graduate School-Rabbit Exhibition Intelligent AIGC Joint Laboratory uses a large amount of structured text data in the legal field for training, and finds senior lawyers to assist manual annotation and conduct high-quality fact-based multi-round dialogue, and at the same time open source three models: ChatLaw-13B, ChatLaw-33B and ChatLaw-TextVec.

The project is committed to providing inclusive legal services, and currently has more than 1k stars on GitHub.

NVIDIA's acquisition of AI startup OmniML is expected to accelerate the use of AI on edge devices

According to the Science and Technology Innovation Board Daily, NVIDIA is further enriching its AI ecosystem through acquisitions, which secretly acquired AI startup OmniML in February. OmniML is a company that specializes in shrinking machine learning models to edge devices, enabling machine learning tasks to be 10x faster on different edge devices with 1/10th the engineering effort.

Last year, it launched a platform called Omnimizer that adapts compute-intensive AI models to low-end hardware, resulting in smaller, faster, and better suited models for the hardware it runs on.

Xueersi Online School launched "Artificial Intelligence First Course", an AIGC course based on self-developed large models

On the afternoon of July 3rd, at the programming product upgrade and artificial intelligence course conference of Xueersi Online School, Xueersi announced the launch of an AIGC (Generative Artificial Intelligence) course based on self-developed large language models, named "Artificial Intelligence First Course". According to the official statement of Xueersi, the course will "promote the upgrading of programming course products and help popularize AI education."

Key Papers:

ChatGPT for Robotics: Design principles and model capabilities

This paper presents experimental research on robotics applications using OpenAI's ChatGPT. We outline a strategy that combines the design principles of prompt engineering and the creation of high-level libraries so that ChatGPT can adapt to different tasks, simulators, and form factors. The research focuses on evaluating the effectiveness of different cue engineering techniques and dialogue strategies for performing various robotic tasks. Studies have shown that ChatGPT can effectively solve some of these tasks while allowing users to interact with it primarily through natural language instructions. In addition to these studies, the research team introduced an open-source research tool, PromptCraft, which includes a platform that allows researchers to jointly upload and vote on excellent bot app prompt schemes, and a sample bot simulator that includes ChatGPT integration to make it easier for users to start using ChatGPT for bot applications.