Early this morning, domestic AI application developers received a notification from OpenAI, mentioning that unsupported countries and regions will be denied access to OpenAI's API, effective from July 9th.

Recently, there were rumors that ByteDance was collaborating with Broadcom to develop advanced AI processors to ensure stable supply of high-end AI chips. In response to this news, ByteDance stated that the information is inaccurate.

What other hot topics in the AI industry at home and abroad deserve attention in the past day? Let's take a look with Crow News.

/ 01 / Large Models

1) OpenAI Announces Blocking API Access in China

Domestic AI application developers received a notification from OpenAI, mentioning that unsupported countries and regions will be denied access to OpenAI's API, effective from July 9th. Those who wish to continue using it can contact relevant services in supported countries and regions.

It is reported that China could not directly access OpenAI's API, and those who originally used local proxies need to change their approach.

2) Apple Shelved Plans to Integrate Meta AI Model Due to Privacy Concerns

According to Bloomberg, Apple had negotiated with Meta several months ago to integrate its AI model into the iPhone but ultimately decided to abandon the plan due to privacy concerns.

3) Release of Streamer-Sales, a Large Model for Live Streaming Sales

Streamer-Sales is an AI model based on InternLM2 for fine-tuning instructions for live streaming sales. It can automatically generate product copywriting and answer questions in real-time. The model features one-click generation of anchor copywriting, reasoning acceleration technology, RAG technology for precise copywriting generation, voice processing capabilities, digital human video generation, and real-time information inquiry. The project is open-source.

4) iFLYTEK to Release Xinhuo Large Model V4.0 This Thursday, Targeting GPT-4 Turbo

iFLYTEK announced that it will release the latest progress of its Xinhuo Large Model on June 27th. The announcement stated that the Xinhuo Large Model has comprehensively enhanced seven core capabilities of the large model base, targeting GPT-4 Turbo, and releasing multiple new products and applications, including iFLYTEK Xinhuo APP/Desk, Xinhuo Intelligent Grading Machine, iFLYTEK AI Learning Machine, iFLYTEK XiaoYi APP, and Xinhuo Enterprise Intelligence Platform.

5) LibreChat, a Free and Open-Source Clone of ChatGPT

LibreChat announced its free and open-source status. It is an enhanced and hosted clone of ChatGPT, supporting multiple AI models and providing flexible customization options, solving the issues of integration and switching between multiple AI models.

6) Robam Appliances Launches the "God of Food" Large Model for Cross-Device Collaboration

Robam Appliances recently released the "God of Food" large model, which provides personalized and tailored solutions for everyone. It not only tells users how to cook but also mobilizes all devices for cross-device collaboration and planning, assisting users in cooking.

/ 02 / AI Applications

1) Alibaba Launches Diffutoon, a Video-to-Animation Tool

Diffutoon is an AI tool developed in collaboration between Alibaba and East China Normal University that can convert realistic videos into various anime styles. It uses advanced AI image generation technology to process videos through diffusion models, preserving details and converting them into cartoon animations, improving processing speed.

2) Synthesia Launches New Version 2.0, Introducing Interactive AI Videos and Full-Body Virtual Characters

Synthesia, a UK startup, has launched a new version of Synthesia 2.0, introducing interactive videos and full-body virtual characters, providing a comprehensive suite for enterprises to accelerate the implementation of video-based communication initiatives. New features include AI screen recording tools to simplify the process of creating operational guides for enterprises. The updated virtual characters have more body language, enhancing the narrative ability of digital characters.

3) Record Labels Jointly Sue AI Music Generation Companies Suno and Udio for Copyright Infringement

Multiple well-known record labels jointly sued two AI startups, Suno and Udio, accusing them of using copyrighted music without permission to train music generation models, infringing on the copyrights of commercial works, and demanding that the defendants承认 infringement and shut down AI services.

4) Yao Class Geniuses Team Up to Develop the Sequel to "I'm Surrounded by Big Models!"

Fan Haoqiang, General Manager of Megvii's Research Department, led the development of the mini-game "I Ruined the Big Models," the sequel to the popular mini-game "I'm Surrounded by Big Models!" with over 10,000 users in a single day. Renowned as a genius youth for his achievements such as winning the IOI gold medal and being admitted to Tsinghua's Yao Class, Fan Haoqiang received technical support from the star AI model entrepreneur StarLeap in this development, providing multi-modal and multi-agent support for large models.

5) EvTexture: Quickly Enhancing Blurry Videos with AI

EvTexture is an event-based visual-driven video super-resolution (VSR) technology that captures high-frequency dynamic details, enhancing the texture quality of videos and making each frame clearer and more realistic.

/ 03 / Investment and Financing News

1) OpenAI Acquires Remote Collaboration Platform Multi Founded by Chinese Google Engineers

OpenAI acquired the remote collaboration platform Multi (formerly Remotion). Multi supports real-time collaboration for teams of up to 10 people through screen sharing, including voice, video calls, and annotations.

Founded by former Dropbox Product Manager Alexander Embiricos (Co-founder & CEO) and former Google engineer Charley Ho (Co-founder & CTO), Multi's co-founders revealed that five members of the company will join OpenAI after the transaction is completed, and Multi will shut down after July 24th.

2) DreamTech Completes Two Consecutive Rounds of Financing Totaling Several Million Dollars

Domestic native 3D generation AI company DreamTech has continuously completed several million dollars in angel round and angel+ round financing. The angel round was led by Origin Capital, with participation from Qiming Venture Partners and Cloud Angel Fund, while the angel+ round was exclusively invested by Early Bird Capital.

The company began official operations in December 2023. CEO Dr. Zhang Feihu graduated from Oxford University, and the founding members of the team include academicians from the British Academies, national-level young talents, and founding members of Tencent Meeting.

3) AI Legal Startup Leya Secures 70 Top Law Firm Partnerships in Its First Year and Raises Funding from Benchmark

Leya, a startup that simplifies legal industry operations with AI tools, has been adopted by over 70 top European law firms, with clients spanning the UK, Spain, France, Portugal, the Netherlands, and the entire Nordic region. Leya currently supports authoritative information from over 15 jurisdictions.

Founded in Sweden in June 2023, Leya received support from Y Combinator in its early stages and successfully completed a $10.5 million seed round led by Benchmark in May 2024.

/ 04 / AI Infrastructure

1) ByteDance Jointly Developing Dedicated Chips with Broadcom? Official Response: Inaccurate Information

Recently, there were rumors that ByteDance was collaborating with Broadcom to develop advanced AI processors to ensure stable supply of high-end AI chips. In response to this news, ByteDance stated that the information is inaccurate. Previously, media reported that the processor developed by ByteDance and Broadcom was a 5nm chip, expected to comply with new US export control regulations, with manufacturing work partially outsourced to TSMC.

2) Musk's xAI Collaborates with NVIDIA, Dell to Build an AI Supercomputer

Elon Musk's AI startup xAI announced a collaboration with Dell, NVIDIA, and Supermicro to build what is said to be the world's largest supercomputer. The project will provide significant computing resources to the startup, reducing reliance on cloud services.

3) Tencent Opens Source New Video Control Method MOFA-Video, Allowing Video Motion Control via Arrows

MOFA-Video is Tencent's latest open-source video control technology, enabling users to control the motion direction of video content through arrows. Simultaneously, MOFA-Video can also migrate facial expressions to newly generated facial videos, creating new characters and emotional expressions.

4) Japan Develops Humanoid Robot Alter3 Using GPT-4 Technology

Alter3, a humanoid robot system developed in collaboration between the University of Tokyo and Alternative Machine, utilizes GPT-4 technology to directly map natural language instructions to robot actions. This research result demonstrates the potential of combining foundation models with robot systems, advancing robot research, although it has not yet been commercialized. Alter3 has demonstrated good action planning and execution capabilities in multiple tests, including daily actions and imitation actions.

5) Dialogue with Lin Yonghua, Chief Engineer of Zhiyuan AI Institute: In-depth Discussion on the Systemic Challenges Behind Open-Source and Closed-Source Large Models

Lin Yonghua, Deputy Dean and Chief Engineer of Zhiyuan AI Institute, believes that open-source and closed-source are not criteria for judging the quality of large models, and that model capabilities depend on algorithms, data quality, and computing power investment. She also pointed out that in the era of large models, model files are more important than source code, and open-source models are easier to evaluate comprehensively because closed-source models can only access APIs and cannot know the details behind them.

6) Ant Group Establishes Tianxiang Security Lab, Focusing on Large Model Infrastructure and Web3 Security

Ant Group announced the establishment of the Tianxiang Security Lab and organized its first expert workshop. The lab will focus on the security of digital infrastructure behind new technologies and scenarios such as large models and Web3, paying attention to talent development, scenario empowerment, ecological openness, and the integration and implementation of industry-university-research technology. Simultaneously, the lab has established an expert advisory committee, with seven experts from universities such as the University of Electronic Science and Technology of China and Huazhong University of Science and Technology serving as committee members.