Sibich DFM-2 Large Model Officially Launched to Support Smart Industry Development

The conversational AI platform SiBiChi has announced that its self-developed large model, DFM-2, has successfully completed its filing and is now officially open to the entire society. The Dongfeng large model is an industry-specific language model with general intelligence, integrating with its self-developed full-chain intelligent dialogue platform, DUI, and has been implemented in areas such as intelligent vehicles, smart homes, government services, healthcare, scientific research, and smart office. The DFM-2 large model, developed by SiBiChi, has passed the filing process and possesses capabilities such as general intelligence and knowledge, addressing industry challenges, and large model technology linkage. Combined with comprehensive full-chain conversational AI technology, the DFM-2 model can meet personalized needs in scenarios such as intelligent vehicles, smart homes, consumer electronics, and industries like finance, rail transit, and government services. The opening of SiBiChi's DFM-2 large model will contribute to the development of the intelligent industry, providing users with higher quality and more efficient services.

Zhipu Launches GLM-5.3-FlashX: Speed Reaches 200 Tokens/s, Domestic Computing Power Advances Again

Zhipu AI launches GLM-5.3-FlashX, with the API also going live, offering a maximum output of 200 tokens/s. It focuses on intelligence, price, and speed, providing high-throughput, low-latency inference for enterprise developers. The predecessor, GLM-5.3-Flash, was previously introduced overseas under the name Ox Alpha. It gained popularity due to its strong intelligence and cost-effectiveness at the same size, with increasing usage volume. Zhipu is now supporting growing demand.

SpaceX AI Department Targets Bankrupt Startups' Data: To Be Used for Training Grok

It has been disclosed that SpaceX's AI department, SpaceXAI, is internally discussing the potential acquisition of customer and operational data from struggling or bankrupt startups to train models such as Grok, in order to obtain high-quality external data to enhance performance. This move indicates that it no longer solely relies on Elon Musk's X platform and internal AI mentor data, but is beginning to expand diverse data sources for Grok.

WeChat Responds to Privacy Controversy Surrounding Weixiao AI: It Will Not Read Users' Deleted Moments

The WeChat AI assistant 'Weixiao' has sparked controversy over privacy concerns and has topped the hot search list. WeChat clarified that claims about directly reading chat records or spying on privacy are inaccurate. The assistant is a native AI, currently in a limited beta test, supporting text and voice interaction, capable of checking weather, express delivery, and takeout. It can also proactively trigger in scenarios such as Moments, official accounts, chats, and short video sections to assist with content summarization.

Gates on the Global AI Race: Four Models in China and the US Lead, the Humanoid Robot Competition in China is Very Impressive

The Gates Foundation will invest $1 billion over the next two years to promote the development of artificial intelligence, making cutting-edge technology benefit the poorest regions. Gates also shared his views on the global large model landscape, computing power carbon emissions, and Chinese humanoid robots, pointing out that according to different measures, there are currently four leading large model companies in both the US and China.

AntGroup Open Sources SingProbe: Real-Time Risk Blocking for Large Model Generation with Lightweight Probe, Computing Overhead Below 0.5%

Ant Group's AI Security Lab opens the internal security guardrail technology called SingProbe. Unlike external review, it reuses the hidden states of the base model for inference, achieving real-time token-level risk assessment with less than 0.5% additional computing power. It can simultaneously perform intent classification, safety detection, and hallucination recognition, and can block risks at the millisecond level before output in high-risk scenarios such as healthcare. It has been adapted to 29 mainstream large models.