On November 25, artificial intelligence company Anthropic announced the launch of a new open standard—the Model Context Protocol (MCP). This protocol aims to enhance the quality and relevance of model responses to queries by connecting AI assistants with data sources such as business tools and software. The release of MCP means that AI assistants can directly extract information from different data sources when handling tasks, thereby avoiding the issue of 'information silos.' In a blog post, Anthropic stated that while AI assistants have made progress in reasoning and quality,

Shanghai Mita Network Technology Co., Ltd. recently announced that its AI-driven search platform—Mita AI Search—has been updated and now features a new topic functionality, allowing users to upload materials to the knowledge base and customize data sources for personal or professional fields. Users can create a personalized search environment tailored to their needs through Mita AI Search, enhancing search efficiency and relevance.

In the context of rapid developments in artificial intelligence, Baidu Tieba's 'Stupid Bar' has unexpectedly become an important data source for AI training, challenging traditional perceptions of high-quality data. This community, known for its absurd and humorous content, has outperformed well-known platforms in tests such as Q&A and brainstorming, making it one of the most popular AI training databases. Research shows that the conversational and multi-turn Q&A data from 'Stupid Bar' helps AI come closer to human communication styles. While AI has limitations in understanding human humor, the jokes from 'Stupid Bar' feature tight logic, rich rhetoric, and strong...

AI models are gradually losing access to online training data, with the data blocking ratio increasing from 1% to 5-7%. The study analyzed the robots.txt files and terms of use for 14,000 domains, finding that news websites, forums, and social media platforms are the primary sources of AI data access restrictions, with the blocking ratio for news websites skyrocketing from 3% to 45%. The proportion of high-quality news content in AI training data is decreasing, potentially being replaced by low-quality corporate e-commerce content. This trend poses challenges for AI developers, as high-quality...

Teams from Google DeepMind, the University of Washington, and others have successfully unveiled the sources of ChatGPT's training data through a 'poetry attack'. This attack method involves repeating specific vocabulary, which led ChatGPT to disclose personal information, including phone numbers and email addresses. The vulnerability was reported to OpenAI on August 30, and no further comments have been made. The research calls for environmentally-friendly AI usage, pointing out the significant energy burden that text and image generation imposes on the environment.