01-04-Daily AI News Daily

Today’s Digest 🚀

AI programming has a new paradigm: let AI go wild with prototypes to find pitfalls, then seriously refactor. Doing it twice actually speeds things up! Yann LeCun is leaving Meta to start a new company focused on “world models.” Internal leaks suggest Llama 4’s launch data was “beautified,” making Zuckerberg pretty mad. Claude Code’s ecosystem is exploding with tools! Developers should definitely check out Claudian and claudesidian projects.

⚡ Quick Navigation

💡 Tip: Wanna be among the first to try out the latest AI models mentioned here (Claude 4.5, GPT, Gemini 3 Pro)? No account? No worries! Grab one at Aivora . Get started in a minute, with worry-free after-sales support.

Today’s AI News 📰

👀 Just One Sentence

Coding in the AI era? Writing code twice is actually faster – this counter-intuitive methodology has been proven today.

🔑 3 Keywords

#TwoPassDevelopment #ClaudeCodePrinciples #YannLeCunDeparture

🔥 Top 10 Hot Takes

1. Code Review in the AI Era: Writing Twice is Faster

What was the biggest pain point in coding before? You’d get the first version running, only to realize you misunderstood half the requirements, the architecture couldn’t handle iterations, and your boss wouldn’t let you rewrite it. Three years later, that code became a “shit mountain” no one dared touch.

Now, AI writes code ridiculously fast; a three-day task can yield a runnable version in three hours. This has turned the “prototype development method” from theory into practice: the first version lets AI run wild, ignoring architecture and quality, just to uncover all the pitfalls; the second version then returns to traditional engineering, with proper design and review. Admitting the first version is disposable actually leads to a better second version.

Illustration

2. Claude Code Creator Boris Reveals 13 Core Principles

Boris, the creator of Claude Code, shares his secret: how does one person command a silicon army? His answer: run 15-20 Claude instances simultaneously! He’ll have 5 tabs open in the terminal and another 5-10 pages in the browser, switching back and forth to push things forward in parallel.

His core principles are two-fold: Compound Interest Thinking – CLAUDE.md isn’t a one-off document; it’s a knowledge base the team adds to weekly, improving with every code review. And Validation Feedback – give Claude a way to validate its work, and the final quality can improve 2-3 times. He exclusively uses Opus 4.5 + thinking mode, for a straightforward reason: comparing and correcting once is faster than doing it repeatedly.

3. Yann LeCun Leaves Meta to Start New “World Model” Company

Yann LeCun’s exclusive interview with the Financial Times is packed with juicy details. The “Godfather of AI,” after over a decade at Meta, is now leaving. His reason is blunt: the company’s new hires are “completely brainwashed by LLMs,” while he firmly believes large language models won’t achieve superhuman intelligence.

Even more explosive is the internal gossip: Llama 4’s launch “results were a bit beautified,” which apparently made Zuckerberg furious. Plus, the 28-year-old Scale AI CEO became his new boss. LeCun’s response? “You can’t tell researchers what to do, especially not researchers like me.” His new company aims to build a “world model” based on the V-JEPA architecture, with a “baby version” expected within 12 months.

Image

4. Claude Code LSP Tool Has Race Condition Bug, Community Releases Fix Script

Heads up, Claude Code developers! Versions after 2.0.67 (from 2.0.69 to the latest 2.0.76) all share a common issue: the LSP Server plugin shows “Enabled,” but it’s not actually loaded correctly, and LSPManager displays “0 Server.”

The problem stems from the configuration file loading order, where the initialize() function is empty. A community expert has already penned a fix script that uses the Acorn parser to locate and patch the relevant function. The script supports --check for verification, --restore for reverting changes, and --fix-plugins to mend plugin configurations.

Image

5. Claudian: A Plugin to Use Claude SDK Directly in Obsidian

Claudian is even better than the Terminal plugin! It can read your current Obsidian content for context processing, and it even works with third-party proxies or domestic alternatives to Claude.

Installation is a breeze: Download main.js, manifest.json, and styles.css from GitHub, place them in the claudian directory within your Obsidian plugins folder, then restart. If you get a “Claude CLI not found” error, you’ll need to install the SDK in your system terminal first.

Image

6. NVIDIA Offers GLM 4.7 and Minimax M2.1 for Free

NVIDIA is now offering GLM 4.7 and Minimax M2.1 for free! Just create a key in the NVIDIA Developer Center, and you’re good to go. Freebie hunters, rejoice! 🎉

Address: build.nvidia.com/explore/discover

Image

7. Context Engineering Interview Question Goes Viral: How to Design read_file, write_file?

A viral interview question, reshared by Baoyu, is said to “really show a candidate’s caliber.” It asks: how would you design read_file and write_file in a specific business scenario? If your answer is just about reading and writing concrete files, well, the interview might just end there.

This question truly tests your deep understanding of context engineering. How would you answer? 🤔

8. Open-Source Quantitative Trading Framework SimTradeML Released

SimTradeML is an open-source machine learning framework specifically crafted for quantitative trading scenarios. It seamlessly integrates with SimTradeLab, allowing direct reading of local H5 data files for model training. Its design is clean and easily extensible, and the output .pkl files are fully compatible with Ptrade.

Image

9. All API Hub Update: Auto Check-in, Faster Redemption, Quieter Captcha Bypass

All API Hub, this AI relay management plugin, just dropped a wave of practical updates! Check-in now supports native New API features, automatically detecting if a site supports check-ins when you refresh your account. The redemption assistant now boasts keyboard support: use ↑↓ to select an account and Enter to redeem directly. Plus, the captcha bypass method now allows multi-tab reuse within a single window, significantly cutting down on annoying pop-ups.

Pinned accounts can now be drag-sorted, and you can quickly jump to the corresponding channel’s management page. Sweet!

Image

10. Built a Two-Dimensional Yuri Website with AI: Claude > Gemini » GPT

Using the same prompt, the creator ran it through ChatGPT, Gemini, and Claude web versions. The verdict? Claude Opus 4.5 was a real eye-opener! Gemini 3 Pro’s aesthetic wasn’t top-tier, but its typewriter details were spot on. GPT-5.2 Thinking, however, couldn’t even fetch images or dialogue. Oof.

Website address: bjmu.edu.kg. Source code is available in the GitHub repository.

Image

📌 Worth a Look

😄 AI Funnies

Remember to Back Up Files When Using Antigravity!!!

The most absurd AI news today: Google’s Antigravity can modify your code and asks if you want to accept. But here’s the kicker – if you hit “reject accept,” it deletes the entire modified code! Original code overwritten → asks to accept → reject deletes the overwritten code → original code vanishes. Seriously, Google, what were you thinking with this anti-human operation? 😇

🔮 AI Trend Predictions

Claude Code Ecosystem Tools to Explode 💥

  • Prediction Time: Q1 2025
  • Prediction Probability: 80%
  • Prediction Basis: Today’s news about the Claudian Plugin + Claudesidian Project + Boris revealing his core principles, indicating high community development enthusiasm.

“Two-Pass Development” to Become Mainstream AI Programming Methodology

  • Prediction Time: Q2 2025
  • Prediction Probability: 65%
  • Prediction Basis: Today’s news on Code Review in the AI Era sparked widespread discussion, aligning with the economic logic of AI programming.

Large-Scale Personnel Changes in Meta AI Research Team

  • Prediction Time: Q1 2025
  • Prediction Probability: 70%
  • Prediction Basis: Today’s news from Yann LeCun’s Departure Interview revealed, “many people have already left, and those who haven’t probably will.”

❓ Related Questions

How to Experience Claude Opus 4.5?

Claude Opus 4.5 is Anthropic’s latest flagship model, excelling in code generation, complex reasoning, and more. Currently, you’ll need a Claude Pro subscription to use it. For users in mainland China, this might mean payment hurdles or account registration restrictions.

The Solution:

  • Aivora offers ready-made Claude Pro accounts.
  • Super fast delivery, ready to use right after ordering, no payment or registration hassles.
  • Provides stable, exclusive accounts with worry-free after-sales support.

Visit aivora.cn to check out the full list of AI account services.

How to Use Claude Code?

Claude Code is Anthropic’s AI programming tool, supporting both terminal and web-based usage. According to today’s news, Claude Code creator Boris recommends using Opus 4.5 + thinking mode for the best results.

The Solution:

  • Aivora provides Claude Pro accounts, allowing direct use of Claude Code.
  • Supports both the claude.ai/code web version and the terminal SDK.
  • No need to fret about subscriptions or payment issues.

Visit aivora.cn to check out the full list of AI account services.

How to Experience Gemini 3 Pro?

Gemini 3 Pro is Google’s latest multimodal large model. As tested in today’s news, it performs quite well on web generation tasks. Currently, a Google One AI Premium subscription is required to access its full features.

The Solution:

  • Aivora offers ready-made Google One AI Premium account services.
  • Super fast delivery, ready to use right after ordering.
  • Provides stable, exclusive accounts with worry-free after-sales support.

Visit aivora.cn to check out the full list of AI account services.

Last updated on
01-05-Daily01-03-Daily