06-15-Daily

AI Insights Daily 2025/6/15

AI Product and Feature Updates

1. In the AI math practice test after the 2025 National College Entrance Examination (Gaokao), the Quark large model topped the charts with excellent scores of 145 and 146, surpassing competitors like Doubao and Yuanbao, setting a new benchmark for domestic AI math capabilities. It not only demonstrated amazing accuracy, but also had a significantly faster answering speed, and its powerful science problem-solving ability has opened a new chapter of heuristic learning for users.
   
   

AI Cutting-Edge Research

1. orange.ai’s tweet revealed a funny story: Someone directly asked Claude Opus to “sign” as the first author and write a short article titled “The Illusion of the Illusion of Thinking,” which was basically a direct “clap back” at Apple’s paper “The Illusion of Thinking” that questioned the reasoning ability of large models, and also “roasted” Apple’s AI research level. This move not only hinted at Claude Opus’s powerful strength in the AI field, but also sparked a philosophical debate about whether large models have the essence of thinking. ‘More Details’
   
   
2. orange.ai brilliantly revealed a “battle of the gods” between Anthropic (Claude) and Cognition (Devin) around the pros and cons of multi-agent systems: Claude strongly supports collective intelligence, believing that multi-agents can break through the context bottleneck of single agents with diversity, and performance can be improved by more than 90%; while Devin poured cold water, warning that multi-agents may cause context inconsistency, information fragmentation, and communication problems. This debate is like a mirror, reflecting the complexity of AI architecture design as comparable to managing a large company. At the same time, it may also foreshadow that after the Scaling Law gradually slows down, the collective intelligence formed by multi-agents will become a key “seedling” for promoting exponential growth in AI. ‘More Details’

AI Industry Outlook and Social Impact

1. Gartner boldly predicts that by 2028, as much as 80% of generative AI commercial applications will be directly incubated on existing data management platforms, which is basically hitting the “acceleration button” for developers, and is expected to shorten project delivery time by half and greatly reduce development difficulty. Among them, Retrieval-Augmented Generation (RAG) technology is regarded as a core weapon, which can make AI models more accurate and reliable, and can also combine the latest enterprise data to inject powerful power into process optimization, user experience improvement, and future insight prediction.
   
   
2. Match Group’s latest research reveals an intriguing new trend: AI companions are quietly becoming a new emotional choice for people. The survey found that 16% of respondents even regard robots as “romantic partners,” and more surprisingly, up to 60% of people believe that having an AI girlfriend or boyfriend does not constitute cheating, which undoubtedly challenges our traditional definition of intimate relationships. However, although AI companions can provide emotional comfort, experts also warn of their potential risks, such as possibly exacerbating social isolation and triggering privacy and ethical issues. This undoubtedly prompts us to deeply reflect on how the future of technology and human emotion will intertwine.
   
   
3. Liko exclaimed that with Cursor and Claude code, these two magical tools, the traditional engineering development method is simply undergoing a “major revolution”! He pointed out that small teams can use the agile collaboration of AI Agents to achieve efficiency that can leave the rigid processes of large companies far behind. The accelerated iteration capabilities of this AI tool can be seen from the Lovable activities and the rapid development practice of the Cursor/Claude team’s own products, which indicates that future innovation will explode at a speed you can’t imagine, and may even make us “wage slaves” feel a sense of “nothing to do”. ‘More Details’
   
   
   
   

Open Source TOP Projects

1. Tencent announced at the CVPR 2025 conference that the Hunyuan 3D 2.1 large model is officially open source! As the first full-link industrial-grade 3D generation large model, it has achieved significant breakthroughs in 3D effects and material performance. Even more exciting is that it even supports consumer-grade graphics card deployment, which greatly reduces the threshold for 3D content creation for ordinary users and developers. This model provides efficient solutions for industries such as games and movies, and has accumulated more than 1.8 million downloads on the Hugging Face platform, which shows its high popularity among global developers. ‘Project Address’
   
   

Social Media Sharing

1. Twitter user wwwgoubuli shared his “advanced” experience of chatting with AI. He found that AI is particularly good at outputting correct and complex long sentences, which brings him a different kind of reading enjoyment. He humorously pointed out that although we usually use short sentences in daily communication, only when we talk to AI can we fully immerse ourselves in the context built by long sentences and full of rich semantic experience. ‘More Details’
2. ginobefun sincerely shared a “hidden gem”: a curated list of AI-related RSS subscriptions that he spent a day organizing, which includes more than 200 technical articles, more than 30 AI podcasts, and more than 150 core AI users on Twitter. It’s basically a “secret manual” for chasing AI trends! He especially recommends using @follow_app_ to import these resources, and praised the AI summarization, translation and recent reader functions it provides, which greatly improves the user experience. ‘Project Address’
   
   
3. Li Jigan shared his unique insights on how to use AI on social media. He pointed out that whether it is the initial “human is fiercer than AI” mode of “I’m the boss” (human-centered), or the “AI is the boss, I’m the servant” mode (vibe coding) that many people mistakenly believe is the way to go, both have limitations. And now he firmly believes that only “human-AI collaborative creation” can truly unlock the potential of AI and maximize the value of technology. ‘More Details’

Listen to the Audio Version

🎙️ Xiaoyuzhou	📹 Douyin
Laisheng Tavern	Laisheng Intelligence Station