Hextra-AI-Insight-Daily/content/en/_index.md

linkTitle, title, breadcrumbs, next, description, cascade

linkTitle	title	breadcrumbs	next	description	cascade
Today's Daily	Today's Daily-AI日报	false	/en/2025-06/2025-06-29	Daily selection of AI industry news, open source hot spots, academic frontiers and big V opinions. AI information; AI daily; AI knowledge base; AI tutorials; AI information daily; AI tools;CMU and Xiaohongshu teams have teamed up to introduce an innovative technique: HoPE (Hybrid of Position Embedding)! 🚀 They found that existing multimodal RoPE struggles a bit with "long-context semantic modeling." So, HoPE cleverly introduces zero-frequency temporal modeling and dynamic scaling s...	type docs

AI Insights Daily 2025/6/30

AI Daily | 8 AM Update | Web-wide Data Aggregation | Cutting-Edge Scientific Exploration | Industry Voices Speaking Freely | Open Source Innovation Power | AI and the Future of Humanity | Visit Web Version ↗️

AI Content Summary

CMU and others push HoPE to enhance VLM long-video understanding; Renmin University and others optimize multimodal models with MokA.
Open-source projects cover generative AI tutorials and AI tool libraries. Gary Marcus questions whether pure LLMs can achieve AGI.
AI significantly lowers the barrier to entrepreneurship, prompting a shift in investment thinking, and encouraging embracing collaboration to seize generational opportunities.

Cutting-Edge AI Research

1. CMU and Xiaohongshu teams have teamed up to introduce an innovative technique: HoPE (Hybrid of Position Embedding)! 🚀 They found that existing multimodal RoPE struggles a bit with "long-context semantic modeling." So, HoPE cleverly introduces zero-frequency temporal modeling and dynamic scaling strategies. This is like equipping Vision-Language Models (VLMs) with "long-distance running shoes," significantly boosting their length generalization capability in long video understanding and retrieval tasks, sending them straight to optimal performance! 💡 How cool is that! 'Paper Link' 'Project Link'

2. Mind-blowing! Renmin University of China and Shanghai AI Lab teams have unveiled a new breakthrough: the MokA (Multimodal low-rank Adaptation) method! 🤯 They discovered that when fine-tuning Multimodal Large Language Models (MLLMs), it's easy to lose sight of the balance between single-modal independent modeling and inter-modal interaction. MokA, however, acts like a master balancer. By cleverly combining modality-specific A matrices, cross-modal attention mechanisms, and shared B matrices, it perfectly solves this problem, making the performance of multimodal tasks shoot through the roof! ✨ Absolutely brilliant! 'Paper Link' 'More Details'

Top Open-Source Projects

1. The "generative-ai-for-beginners" project (with 86,547 stars 🌟) has launched 21 lessons specifically designed for beginners, teaching you hands-on how to master the building skills for generative AI! Want to become an AI wizard? Dive in and learn! 💪✨ 'Project Link'

2. The "system-prompts-and-models-of-ai-tools" project (with 62,777 stars ✨) is an absolute treasure trove! It compiles system prompts, tools, and AI models from popular AI tools and agents like Cursor and Devin, offering you a comprehensive, one-stop reference to help you master AI tools! 📚💡 'Project Link'

3. The "storm" project (with 24,892 stars ⭐) is seriously impressive! It's an LLM-powered knowledge management system that can act like a mini-researcher, independently researching specific topics and then generating complete reports with citations. For writing papers or doing research, it's an absolute lifesaver! 🧠✍️ 'Project Link'

Social Media Buzz

1. Renowned AI scholar Gary Marcus has "fired shots" again! 🤔 Citing papers from MIT, the University of Chicago, and Harvard University, he bluntly points out that pure LLMs simply cannot achieve Artificial General Intelligence (AGI)! Why not? Because they suffer from "Potemkin understanding" (a false understanding) and conceptual inconsistency. Simply put, AI might perform brilliantly on tests, but it fumbles when truly understanding and applying concepts. The research also found that LLMs like GPT-4o, after concepts are clearly defined, see their performance plummet 📉 when applied to practical tasks like classification, generation, or editing. They even show conflicting representations internally for the same idea. This has sparked widespread attention and testing from industry bigwigs like Google DeepMind scientist Prateek Jain! Looks like AI still has a long way to go to reach AGI! 💡 'More Details'
   
   

2. Tom Huang has spilled the beans on Cursor's core developers' productivity hacks! 🚀 Want to use Cursor more efficiently? They'll show you how to use "Parallel Agents"! By cleverly combining Tab, Formed Tab, and Background Agent, you can build a super-efficient task execution system that will make your AI collaboration 💻 incredibly powerful! Go check out how to do it! 'More Details'
   
   

3. Teacher Yang Yi has put forward a thought-provoking idea: the content creation field is currently in an "attention arbitrage window" 😮‍💨! He says some folks are already using AI to "build content leverage," hinting that in the future, once AI becomes widespread, human original content will become increasingly valuable, even commanding a premium. But what worries him more is that AI might gradually "erode human spiritual culture" at extremely low costs—and that's far scarier than just a shift in content creation methods! ✍️ Food for thought... 'More Details'

4. Teacher Yang Yi believes that in the AI era, the barrier to entrepreneurship has been shattered by AI! 💸 The cost of building an MVP (Minimum Viable Product) has plummeted, making rapid idea validation possible. His advice to entrepreneurs is: stop overthinking whether an idea is feasible or not; just use AI to validate an MVP in as little as 3 days, or even quickly test 30 ideas in 3 months! This way, you'll find that truly worthwhile direction to pour your heart into much faster! 🚀💡 That's super empowering! 'More Details'

5. As an AI investor, Yang Yi shared his "secret weapon" 📈: he doesn't focus on hard data but rather prioritizes qualitative metrics! He believes there are five key points to determine if an AI startup has investment potential: the founder's grand vision for the future path (including PMF and scalability), the team's unwavering conviction, how much efficiency AI has boosted in team management, whether the Agent has a complete feedback loop (which is the methodology for AI success!), and the scalability of the multi-agent framework. He feels that data like user retention are just "byproducts" that naturally emerge over time! 🎯 Unique insight! 'More Details'

6. A user shared a "new approach" 👨‍💻 for coding with AI, and this mode is becoming increasingly popular: Instead of rushing to give AI detailed instructions, first brief it on the project background and goals, then let the AI propose ideas based on that information, and then together align the granularity and discuss. This approach cleverly leverages AI's efficiency in quickly understanding context, compensating for our human "brain cell deficit" when making detailed plans, and significantly boosting workflow efficiency in a collaborative mode! 🤝 It's a godsend for programmers! 'More Details'

7. A user complained that some investors nowadays are still using the outdated data metrics from the mobile internet era to evaluate AI projects, and the result is—they simply can't find good projects! 🤔 Because those traditional logics (formal, informal, and even probability theory) are all about looking back at the past. However, the author emphasizes that Bayes' Theorem is truly a future-oriented decision-making method, better suited for making investment judgments for projects in the AI industry! 💡 It's time to update your investment "operating system"! 'More Details'
   
   
   
   

8. Dashuai Laoyuan and his colleague Dash bluntly point out: the emergence of AI has literally "leveled the playing field" 🏃‍♀️💨 for all of humanity! They believe that the enormous opportunities brought by AI even surpass the internet wave of 20 years ago, enabling everyone, including junior employees, to break free from resource limitations and fully leverage AI for learning and creation. But they also warn that if programmers remain set in their ways and don't strive for progress, that "starting line" will eventually catch up to you and even leave you behind! So, actively embracing AI is the way to go!

---

Listen to the Audio Version of AI Daily

🎙️ Xiaoyuzhou	📹 Douyin
Laisheng's Tavern	Laisheng's Intel Station