01-23-Daily AI News Daily

Daily Summary

Zhipu GLM-4.7 has cracked YouTube video auto-editing Skills—download, translate, burn subtitles all in one go. The lazy creator's secret weapon is now open source.

CodePlan data shows domestic model token usage has surpassed Claude series. Real-world coding performance is verified, and the supply chain crisis is actually accelerating the replacement process.

Live Skills tutorial tonight at 8 PM—don't miss your chance to jump in.

⚡ Quick Navigation

💡 Tip: Want to experience the latest AI models mentioned in this article (Claude 4.5, GPT, Gemini 3 Pro) right away? No account? Grab one at Aivora —one minute setup, hassle-free support.

Today’s AI News

👀 One-Liner

Claude Code’s Skills tutorial goes live tonight with expert instruction, while domestic models are quietly becoming the new favorite replacing Claude.

🔑 3 Keywords

#SkillsEvolution #DomesticRise #VideoEditingGameChanger

🔥 Top 10 Headlines

1. One-Click YouTube Video Editing into Bilingual Shorts with Skills

Used to spend half a day editing a YouTube video? This Skills tool now auto-downloads videos, extracts subtitles, intelligently segments, translates bilingual captions, burns them into video, and generates social media copy. The craziest part? Tested with Zhipu GLM-4.7, done in one shot, nearly zero cost. Isn’t this the “lazy creator’s dream tool”? The project is open source, supports Claude Code, and you can swap in domestic models to save money.

Image

2. CodePlan Users Are Already Replacing Claude with Domestic Models

When Claude hit supply chain crisis, CodePlan users didn’t panic. Over the past two weeks, token consumption from GPT/Gemini + domestic open-source models has far exceeded Claude series, with daily usage still climbing. What does this tell us? Domestic models can now handle real-world coding scenarios. CodePlan’s cross-compatible API protocol conversion lets users seamlessly switch to other models when Claude runs out, zero refunds, minimal complaints. Anthropic, you won—but we didn’t lose.

3. TELOS Framework: The “North Star” That Makes AI Actually Understand You

AI always gives you generic answers? That’s because it doesn’t know who you are. TELOS is a structured self-interview framework that transforms “who you are, what you want” into context AI can understand through 6 layers of questions (problems, goals, challenges, strategies, projects, workflows). The coolest part? It evolves. When you read a new book or learn new tech, AI will compare it against your TELOS and suggest system upgrades. This isn’t making AI smarter—it’s making you clearer.

Image

4. Pro Tips for Making Claude Code Self-Learning and Evolving

After each bug fix, have Claude write the experience into global memory. Since Claude.md loads every time, reflecting on lessons prevents repeat mistakes. But heads up—the official recommendation is keeping global Claude.md under 200 lines, so create subfiles instead and index them in Claude.md, telling it when to check experience logs to avoid wasting tokens. It’s like giving AI an “error notebook”—the more you use it, the smarter it gets.

Image

5. Gemini Now Helps You Prep for SAT—Free

Still paying for SAT prep? Gemini partnered with The Princeton Review to launch free SAT practice tests with instant scoring, clear explanations, and personalized study plans. This is a godsend for students. But it also signals AI shifting from “tool” to “tutor”—education disruption might come faster than we think.

6. Google Paper: Ask the Same Question Twice, Gemini Accuracy Jumps 5x

Sounds like a joke, but it actually works. Google’s research found that repeating your question boosts accuracy for non-reasoning or reasoning-disabled models without affecting length or response speed. What’s the mechanism? Possibly the model better understands context on the second read. That said, this doesn’t help reasoning-heavy questions, but it doesn’t hurt either. Simple, crude, but effective.

7. Baoyu’s Claude Code Tips: Don’t Max Out Context

Context compression is Claude Code’s most frustrating bottleneck. Baoyu shares his playbook: disable auto-compression, save intermediate results to files, do one thing per session, and use Git rollback when stuck. Core principle: chat as memory, files as storage, Git as time machine. This approach keeps you productive in Claude Code without waiting 2 minutes for context compression each time.

Image

8. Pencil.dev’s Prompts and Schemas Got Reverse-Engineered

Pencil.dev is Electron-based, meaning .asar files can be unpacked. So someone extracted its prompts and schema definitions—basically types + comments laid out in JSON. For anyone wanting to learn AI tool design, this is treasure. But it’s also a reminder: Electron app “secrets” aren’t that secret.

Image

9. VidBee: Download Videos from Almost Any Site Globally

Need to download videos but can’t find the tool? VidBee downloads from almost any site worldwide, open source and free. Essential for content creators. Just remember to respect copyright—don’t use it for shady stuff. The project already has 4,565 stars, proving demand is real.

10. Tonight at 8 PM, Guizang and Listenhub Teach You Skills Mastery

Want to learn Skills but don’t know where to start? Tonight at 8 PM, Guizang and Listenhub go live teaching everyone how to master Skills. This is a rare learning opportunity. If you’re curious about Claude Code Skills, don’t miss this stream.

Image

📌 Worth Following

[Products]

[Research]

[Open Source]

[Other]

😄 AI Fun

Baidu Wenxin 5.0 Launches, Claims to Lead GPT-5

Baidu’s back. Yesterday Baidu released Wenxin 5.0, claiming the largest parameters ever (24 trillion)—5x DeepSeek, 2.5x Kimi 2. Pros: fully multimodal. Cons: painfully slow. User comments: “Big parameters, big power, but more people probably means slower.” That said, it handles text Q&A and sentiment tasks decently—feel free to test other areas.

🔮 AI Trend Predictions

Claude Code Skills Ecosystem Will Explode in Q1

  • Prediction Timeline: Q1 2025
  • Confidence: 75%
  • Rationale: Today’s YouTube video editing Skills + recent open-source Skills projects (like Baoyu’s baoyu-skills), plus tonight’s tutorial stream—all signal that Skills barriers are dropping and developer communities are scaling fast.

Domestic Models Will Capture More Coding Market Share Than Claude

  • Prediction Timeline: Q2 2025
  • Confidence: 60%
  • Rationale: Today’s CodePlan users replacing Claude with domestic models + Zhipu GLM-4.7, Minimax 2.1 and other domestic models showing stronger coding performance, combined with Claude’s supply chain instability—users are actively seeking alternatives.

AI Video Editing Tools Will Become Standard for Content Creators

  • Prediction Timeline: Q1 2025
  • Confidence: 80%
  • Rationale: Today’s YouTube video editing Skills + popularity of VidBee and similar download tools show creators desperately need video editing—AI tools are rapidly filling this market gap.

❓ Related Questions

How Do I Experience Claude Code’s Skills Feature?

Claude Code’s Skills feature currently requires a Claude Pro or Team subscription. Domestic users may face payment difficulties or account registration restrictions.

Solution: Visit Aivora to get ready-made accounts—instant delivery, worry-free support.

Last updated on
01-24-Daily01-22-Daily