Weekly#12- Mini Vacation

By Harry / 2025-04-07 / In categories weekly

Weekly

Translations: ZH-CN

Preface

Too Late Life & Records & Thoughts & null from 2025-03-31 to 2025-04-06

The cover photo was taken at the Zhongshan Museum during Qingming Festival
Why not show photos from inside the museum?
Because I didn’t check the time and arrived just as they were closing

How to Block AI Crawlers

Issue 343 of Ruan Yifeng’s Weekly: Tech Enthusiast Weekly (Issue 343): How to Block AI Crawlers

It mentions an article published by the owner of SourceHut, discussing how AI models require vast amounts of data, leading to aggressive crawling behavior that puts significant pressure on servers due to high frequency and volume.

He recommends an alternative approach to Cloudflare: increasing the cost for crawlers to collect data

Using the Anubis project: GitHub - TecharoHQ/anubis: Weighs the soul of incoming HTTP …

This tool is interesting - it makes users load JavaScript files containing proof-of-work algorithms that perform intensive calculations.

It effectively consumes a crawler’s CPU resources.

The evidence proves its effectiveness. One site owner reported that over two and a half hours, their website received 81,000 requests, but only 3% passed Anubis’s proof-of-work challenge, suggesting that 97% of the traffic might be bots!
This is insane and shows how rampant AI crawlers have become.
If your website faces similar issues and you can’t use Cloudflare, you might want to try Anubis’s proof-of-work solution.

Haha, I’ll try it when I have time, though it would degrade the browsing experience for regular users.

Google Will Not Open-Source Android Development

https://www.androidauthority.com/google-android-development-aosp-3538503/

CSDN Actually Charges for AI-Generated Answers

https://linux.do/t/topic/524823

CSDN is full of plagiarism and copied content. The platform doesn’t regulate this, creating a poor environment.

Especially in the last two years with the rise of generative AI, they’ve been directly outputting low-quality AI-generated answers to the platform and charging for them… seriously?

Retiring a VPS

I have a Seattle VPS that’s expiring soon, and I’m planning to let it go (not renew)

After using this VPS for two years at a very low price, I’d say it was worth it.

In the final period, using the hysteria2 protocol significantly improved the user experience.

hysteria2 truly is a lifesaver for network connections.

What is a “小鸡” (small chicken)?: It’s a VPS, see reference: VPS

Interesting Project

https://github.com/QIN2DIM/hcaptcha-challenger

hCaptcha Challenger (v0.13.0+) builds an end-to-end Agentic Workflow using the Spatial Chain-of-Thought (SCOT) capabilities of large language models, enabling Agents to follow instructions for spatial visual tasks without additional training or fine-tuning of CNN expert models.
In hCaptcha Challenger, the Agent controls browser pages through Playwright. In your task workflow, the Agent is initialized with a Page object and takes over interactions on the current page. You can use Agent to implement two independent operations: click_checkbox and wait_for_challenge.
hCaptcha was one of the first pioneers to apply image diffusion and synthesis to the CAPTCHA field. With rapid developments in automation engineering, hCaptcha can implement extremely frequent challenge type changes. This has led to increasing difficulties for the community in dealing with frequently updated challenges over the past two years. Traditional convolutional neural networks (CNNs) struggle to achieve good generalization for small datasets in object detection tasks. The complete model fine-tuning process requires significant time and effort, often taking up to half a week to train a CNN model suitable for production. However, by the time model training is complete, hCaptcha may have already updated to new challenge types, causing the freshly trained model to quickly become obsolete or ineffective.
Therefore, the community urgently needs a powerful general visual solution to effectively address spatial visual challenges. Regardless of how frequently hCaptcha updates verification types, this solution can quickly adapt to environmental changes and autonomously control the browser to complete various CAPTCHA tasks without human guidance.

Zhongshan

Visited Zhongshan during the Qingming Festival

A small city

The pigeon dishes were quite good

The next day I went to the Zhongshan Museum (the only place I was interested in), but I didn’t check the time and arrived just as it was closing

Coffee shop next to Zhongshan Museum

Too Late

Input

🎧Podcasts

📚Articles

📚Books

🎥Shows

🎸Music

Harry

Harry

ENFP | Full Stack Engineer | Loves exploring and using new technologies