06-18-Daily

AI Insights Daily 2025/6/18

AI Product and Feature Updates

1. Rokid is teaming up with Alipay to launch the world’s first Rokid Glasses smart glasses and their innovative payment feature, “Look and Pay”! Users can quickly complete payments with just a few words and a scan, which is expected to double efficiency. This smart payment product, which balances convenience, security, and privacy, uses voiceprint multi-factor authentication and real-time risk control, signaling that the future of payment methods will usher in an “eye”-catching showdown, completely changing our consumer experience!
   
   
2. At the recent Baidu AI Day, Baidu unveiled its trump card, successfully creating the industry’s first Luo Yonghao digital human, and announced four key technological breakthroughs in highly persuasive digital humans, vowing to completely revolutionize live streaming marketing and user experience. To popularize digital human live streaming, Baidu has also launched the “Dream Butterfly Plan” and the “Starlight Plan,” with ambitious plans to double the number of top influencer digital humans, and add 100,000 free digital humans and hundreds of millions in subsidies, aiming to enable more ordinary people and small and medium-sized enterprises to easily use digital human live streaming and start a new era of e-commerce!
   
   
3. The Doubao computer and web versions recently officially launched a new “AI Podcast” feature. Users can simply upload files or links to easily generate podcasts in the form of a two-person conversation, which is simply a revolution in the way information is processed and received! This feature not only naturally simulates the spoken language habits of real-life podcasters, but also greatly simplifies the tedious process of content creation and information acquisition, especially in work and study scenarios. It’s a productivity godsend, making knowledge acquisition as easy and fun as listening to a story.
   
   
4. Alibaba Group has launched a major offensive, releasing an upgraded version of the Qwen3 AI model, which is now perfectly adapted to Apple’s MLX architecture. This undoubtedly paves the way for the official launch of Apple Intelligence in the Chinese market, a tailor-made surprise for Apple fans! The new version of Qwen3 not only supports as many as 119 languages and dialects, but also brings a more intelligent and convenient AI experience to the majority of Chinese users with its powerful performance and hybrid reasoning capabilities, making intelligent life within reach.
   
   
5. LinkedIn has comprehensively upgraded its job search experience, launching a revolutionary AI job search feature that completely eliminates rigid keyword restrictions, allowing job seekers to describe their ideal positions in plain language, thereby obtaining more accurate job recommendations! This innovation, based on large language models (LLM), aims to enable every job seeker to find the most suitable job for them more intuitively and efficiently. It’s a total “helping hand” on the job search journey!
   
   
6. Guicang deeply analyzed the video essence of Google’s Gemini team’s product and R&D leader, summarizing the “three axes” of their excellent coding model concept: focusing on data and methodology, codebase context, and Agentic coding, to comprehensively improve programming capabilities. Their ultimate goal is to empower non-professional developers to achieve “Vibe Coding,” making programming as free as creating music. The team firmly believes that “code is everything” is a universal solution tool, always paying attention to real-world value and generalizability, aiming to build an excellent general-purpose model and lead a new wave of programming!
   
   ‘More Details’

AI Frontier Research

1. Tencent’s AI team recently released the AI singing model LeVo. With its amazing zero-shot timbre cloning, stem generation, and high-fidelity music performance, this model can even rival Suno 4.5, the “Siri” of the AI music world, in several key indicators! Tencent has also generously announced that LeVo will be released in open source form, aiming to break down creative barriers and allow more people to easily use AI music, jointly promoting the vigorous development of the AI music ecosystem. In the future, everyone will be a “karaoke king”! ‘More Details’
   
   
2. A recent study revealed an amazing memory leap in large language models: Meta’s latest Llama 3.1 70B model can actually “remember” 42% of the content of the first Harry Potter book, which is nearly ten times the capability of its previous generation model! This milestone not only indicates that AI is rapidly approaching human cognitive levels in terms of deeply understanding and processing text, but also opens up endless possibilities for us to envision the future of AI capabilities - maybe in the future AI can really read all the books for us!
   
   
3. This study proposes a clever method called “budget guidance,” which can effectively control the reasoning length of a large language model without fine-tuning it, as if “limiting” the model’s thinking, thereby significantly reducing reasoning costs while maintaining or even improving performance. The method has shown up to a 26% improvement in accuracy in mathematical benchmark tests, and can effectively reduce the consumption of computing resources. More amazingly, it also has emerging capabilities such as estimating the difficulty of problems, making large models more “cost-effective”! ‘Paper Address’
4. Ego-R1 is a new framework that utilizes the Chain-of-Thought of Tools (CoTT) process and the Ego-R1 agent trained by reinforcement learning to effectively reason about first-person videos lasting for days or even weeks, just like “Sherlock Holmes”. The framework successfully tackles the unique challenge of understanding ultra-long first-person videos, extending the video’s time coverage from a few hours to an amazing week. It’s like giving AI a pair of “never blinking” eyes! ‘Paper Address’

AI Industry Outlook and Social Impact

1. OpenAI recently signed a one-year $200 million contract with the U.S. Department of Defense to develop advanced artificial intelligence tools for the Pentagon in and around Washington, D.C. to address national security challenges, expected to be completed by July 2026. This move not only marks OpenAI’s first collaboration with the U.S. Department of Defense, but also highlights the key role and broad prospects of artificial intelligence in national security strategies. The battlefields of the future may really rely on AI for “strategic planning”!
   
   
2. Wu Bingjian_bj.ai put forward a profound view on the future impact of LLM, cleverly comparing it to the impact of Meitu Xiu Xiu on appearance, predicting that people may become dependent on LLM due to its greatly improved intelligence. This phenomenon prompts us to deeply reflect on the boundaries of human capabilities in the future human-machine symbiosis model - when AI becomes an “intelligence filter,” how will our own wisdom be defined? ‘More Details’

Open Source TOP Projects

1. The “Moonshot AI” team recently released the open source large language model Kimi-Dev-72B, which is simply a boon for programmers, designed to greatly improve programming efficiency and solve code problems! It performs excellently in the SWE-bench Verified test, especially excelling at fixing code defects in the Docker environment. This model is “honed” through reinforcement learning, can accurately locate and solve code problems, and adopts a two-stage framework to simplify the repair process, predicting that software development will become more intelligent and efficient, and the code of the future may be “written” by AI!
   
   
2. The project, named fluentui-system-icons, currently has 7690 stars and provides a series of familiar, friendly, and modern icons, making it an indispensable “material library” for designers and developers! ‘Project Address’
3. Project jan has earned 29967 stars and is a powerful open source alternative to ChatGPT. Its unique feature is that it can run 100% offline on the user’s computer, which is simply a “secret weapon” tailored for users who pursue local privacy protection and control! ‘Project Address’
4. DeepEP is an efficient expert parallel communication library that has received 7795 stars. Its mission is to significantly improve the communication efficiency of related systems like a “network accelerator,” making data transmission lightning fast! ‘Project Address’
5. automatisch is an open source project with 9063 stars that aims to be a free alternative to Zapier, helping users build workflow automation for free and efficiently. The project is committed to solving the time and money cost problems faced by users in the automation construction process, which is simply a boon for small and medium-sized enterprises and individual enthusiasts! ‘Project Address’

Social Media Sharing

1. Yang Yuancheng Koji shared the latest news from the streets of San Francisco, pointing out that a product called “Manus” has appeared prominently on the streets, strongly suggesting that it is actively entering the market and preparing to show its skills! This message is accompanied by two physical images that clearly show the actual existence of Manus in the urban environment, making people full of curiosity about this mysterious product!
   
   ‘More Details’

Listen to the Audio Version

🎙️ Xiaoyuzhou	📹 Douyin
Next Life Tavern	Next Life Intelligence Station