言零_yanling

为积极响应国家关于加强软件工程学科建设与系统安全人才培养的战略部署，特举办软件系统安全赛。本赛事面向大学生，聚焦大型工业软件、关键基础软件等核心领域软件漏洞的挖掘、攻击与修复，通过模拟真实场景的攻防对抗，全面提升大学生在软件系统安全领域的实战能力。
通过此项科技创新活动，有效提高学生的软件系统安全攻防水平、创新意识与团队协作精神，加强高校间的学术交流，推动软件工程与网络安全人才培养体系的深化改革和

---
📊 市场数据：BTC $71,118.26 (-4.32%)
#Crypto #Bitcoin #Binance

引言随着大型语言模型（LLM）和深度学习技术在安全领域的广泛应用，传统的基于特征码、抽象语法树（AST）或沙箱行为分析的 Webshell 检测手段正逐渐向基于 AI 的语义分析演进。AI 模型能够理解代码的逻辑意图，从而识别出经过复杂混淆的恶意脚本。然而，AI 模型本身存在一个本质性的弱点：它们在处理代码时，往往将代码逻辑与注释、元数据等“非执行内容”置于同一语义空间进行推理。这就为提示词注入（

---
📊 市场数据：BTC $71,118.26 (-4.32%)
#Crypto #Bitcoin #Binance

In response to the national strategy to strengthen the construction of the software engineering discipline and the training of system security talents, a software system security competition is specially held. This competition is aimed at university students and focuses on the discovery, attack, and repair of software vulnerabilities in core areas such as large industrial software and critical infrastructure software. Through simulated real-world scenarios of offensive and defensive confrontation, it comprehensively enhances university students' practical abilities in the field of software system security.
Through this technological innovation activity, students' software system security offensive and defensive levels, innovation consciousness, and team collaboration spirit are effectively improved, academic exchanges among universities are strengthened, and the deep reform of the software engineering and cybersecurity talent training system is promoted.

---
📊 Market Data: BTC $71,197.62 (-4.46%)
#Crypto #Bitcoin #Binance

Introduction With the widespread application of large language models (LLMs) and deep learning technology in the field of security, traditional webshell detection methods based on feature codes, abstract syntax trees (AST), or sandbox behavior analysis are gradually evolving towards AI-based semantic analysis. AI models can understand the logical intent of code, thereby identifying malicious scripts that have undergone complex obfuscation. However, AI models themselves have an inherent weakness: when processing code, they often reason about code logic in the same semantic space as 'non-executable content' such as comments and metadata. This creates a vulnerability for prompt injection.

---
📊 Market Data: BTC $71,197.62 (-4.46%)
#Crypto #Bitcoin #Binance

In active response to the national strategy for strengthening the construction of the software engineering discipline and the training of system security talents, a software system security competition is specially held. This competition is aimed at college students, focusing on the exploration, attack, and repair of software vulnerabilities in key areas such as large industrial software and critical infrastructure software. Through simulating real scenario offensive and defensive confrontations, it comprehensively improves college students' practical abilities in the field of software system security.
Through this technological innovation activity, effectively enhance students' offensive and defensive levels in software system security, innovation awareness, and team collaboration spirit, strengthen academic exchanges among universities, and promote the deepening reform of the software engineering and cybersecurity talent training system and

---
📊 Market Data: BTC $71,816.19 (-3.59%)
#Crypto #Bitcoin #Binance

Introduction As large language models (LLMs) and deep learning technologies are widely applied in the security field, traditional webshell detection methods based on feature codes, abstract syntax trees (AST), or sandbox behavior analysis are gradually evolving towards AI-based semantic analysis. AI models can understand the logical intent of code, thus identifying malicious scripts that have been obfuscated in complex ways. However, AI models themselves have an inherent weakness: when processing code, they often reason about code logic alongside "non-executable content" such as comments and metadata in the same semantic space. This creates an opportunity for prompt injection (\n\n---\n📊 Market Data: BTC $71,816.19 (-3.59%)\n#Crypto #Bitcoin #Binance

In response to the national strategy for strengthening the construction of software engineering disciplines and cultivating talent in system security, a software system security competition is being held. This competition is aimed at university students and focuses on the discovery, attack, and repair of software vulnerabilities in core areas such as large industrial software and critical infrastructure software. Through simulating real-world attack and defense scenarios, it comprehensively enhances students' practical abilities in the field of software system security.
Through this technological innovation activity, students' software system security attack and defense levels, innovative consciousness, and team collaboration spirit are effectively improved, academic exchanges between universities are strengthened, and the in-depth reform of the software engineering and network security talent cultivation system is promoted.

---
📊 Market Data: BTC $71,041.52 (-4.74%)
#Crypto #Bitcoin #Binance

Introduction As large language models (LLMs) and deep learning technologies are widely applied in the field of security, traditional methods of Webshell detection based on feature codes, abstract syntax trees (AST), or sandbox behavior analysis are gradually evolving towards AI-based semantic analysis. AI models can understand the logical intent of code, thus identifying malicious scripts that have been obfuscated in complex ways. However, AI models have an inherent weakness: when processing code, they often reason about code logic in the same semantic space as "non-executable content" such as comments and metadata. This creates a vulnerability for prompt injection.

---
📊 Market Data: BTC $71,041.52 (-4.74%)
#Crypto #Bitcoin #Binance

In response to the national strategic deployment to strengthen the discipline construction of software engineering and cultivate talents in system security, a software system security competition is specially held. This competition is aimed at college students and focuses on the excavation, attack, and repair of software vulnerabilities in core areas such as large industrial software and key foundational software. Through simulating real-world attack and defense scenarios, it comprehensively enhances college students' practical abilities in the field of software system security.
Through this technological innovation activity, it effectively improves students' levels of offensive and defensive skills in software system security, innovation awareness, and team collaboration spirit, strengthens academic exchanges among universities, and promotes the deep reform of the talent cultivation system in software engineering and network security.

---
📊 Market Data: BTC $72,961.35 (-1.42%)
#Crypto #Bitcoin #Binance

Introduction As large language models (LLMs) and deep learning technologies are widely applied in the security field, traditional methods of Webshell detection based on feature codes, abstract syntax trees (AST), or sandbox behavior analysis are gradually evolving towards AI-based semantic analysis. AI models can understand the logical intent of code, thereby identifying malicious scripts that have undergone complex obfuscation. However, AI models have an inherent weakness: when processing code, they often infer the logic of the code alongside "non-executable content" such as comments and metadata within the same semantic space. This poses a risk for prompt injection.

---
📊 Market Data: BTC $72,961.35 (-1.42%)
#Crypto #Bitcoin #Binance

In order to actively respond to the national strategic deployment on strengthening the construction of the software engineering discipline and the training of talent in system security, a software system security competition is specially held. This competition is aimed at college students, focusing on the discovery, attack, and repair of software vulnerabilities in core areas such as large industrial software and critical infrastructure software. Through simulating real-world attack and defense scenarios, it comprehensively enhances the practical abilities of college students in the field of software system security.
Through this technological innovation activity, students' levels of software system security offense and defense, innovative awareness, and team spirit are effectively improved, strengthening academic exchanges among universities, and promoting the deep reform of the software engineering and cybersecurity talent training system.

---
📊 Market Data: BTC $74,063.19 (-0.17%)
#Crypto #Bitcoin #Binance

IntroductionWith the widespread application of large language models (LLMs) and deep learning technology in the security field, traditional methods of Webshell detection based on feature codes, abstract syntax trees (AST), or sandbox behavior analysis are gradually evolving towards AI-based semantic analysis. AI models can understand the logical intent of the code, thereby identifying malicious scripts that have undergone complex obfuscation. However, AI models have an inherent weakness: when processing code, they often infer the code logic along with "non-executable content" such as comments and metadata in the same semantic space. This creates a vulnerability for prompt injection.

---
📊 Market Data: BTC $74,063.19 (-0.17%)
#Crypto #Bitcoin #Binance

In response to the national strategic plan for strengthening the construction of software engineering disciplines and training talents in system security, a software system security competition is specially held. This competition is aimed at college students, focusing on the discovery, attack, and repair of software vulnerabilities in key areas such as large industrial software and critical infrastructure software. Through simulating real-world offensive and defensive confrontations, it comprehensively enhances college students' practical capabilities in the field of software system security.
Through this technological innovation activity, it effectively improves students' offensive and defensive levels in software system security, innovation awareness, and teamwork spirit, strengthens academic exchanges among universities, and promotes the deepening reform of the talent training system in software engineering and cybersecurity.

---
📊 Market Data: BTC $73,983.82 (-0.10%)
#Crypto #Bitcoin #Binance

Introduction: With the widespread application of large language models (LLM) and deep learning technology in the security field, traditional webshell detection methods based on feature codes, abstract syntax trees (AST), or sandbox behavior analysis are gradually evolving towards AI-based semantic analysis. AI models can understand the logical intent of code, thus identifying malicious scripts that have undergone complex obfuscation. However, AI models have an intrinsic weakness: when processing code, they often reason about code logic in the same semantic space as "non-executable content" such as comments and metadata. This creates a vulnerability for prompt injection.

---
📊 Market Data: BTC $73,983.82 (-0.10%)
#Crypto #Bitcoin #Binance

1. Introduction

In the past two days, has everyone been bombarded with news about "lobsters" (OpenClaw)?

News about it is everywhere, even representatives at the Two Sessions and Xinhua News Agency are talking about it. It's truly astonishing that an AI software can generate such a huge response.

People are very enthusiastic, the free offline installation events are crowded, and the online "paid installation" business is booming.

Many people might still not know that there is now the simplest way to use lobsters: ArkClaw.

It's so simple that you don't even need to worry about installation because it's a plug-and-play solution that directly comes with lobsters built-in.

I just started using it yesterday and can't wait to share my initial impressions with everyone. Those who haven't used it can treat this as a tutorial on "Getting Started with Lobster with Zero Barriers" to see what lobsters are all about.

2. What is ArkClaw?

Here's the deal, old readers might remember that I reviewed Byte's latest Seed 2.0 model before the Spring Festival.

In the article, I mentioned that this is Byte's strongest foundational model so far; it’s what the mobile Doubao uses and it performed quite well in tests.

The team at Byte later gifted me a Coding Plan package to continue testing this model, and various AI programming tools can call its API (of course, the package also includes other domestic models for free use).

This past Monday, I suddenly discovered that Byte’s Coding Plan package has launched a bundled service, which is ArkClaw.

I asked customer service and found out that as long as you activate the Coding Plan now, you can use lobsters for free.

In other words, as long as you use Byte's AI programming package, you won't need to spend an extra penny. Byte provides a remote host with lobsters already installed for your free use.

It should be noted that the Coding Plan is divided into two packages: lite (first month 9.9 yuan) and Pro (first month 49.9 yuan). The lite package can only be experienced for free for 7 days, and only the Pro package can use ArkClaw for a long time.

3. Cloud Lobster Farming

ArkClaw belongs to "cloud lobster farming" (also known as "cloud lobsters"), which means installing lobsters (OpenClaw) on the cloud host of Volcano Ark (Byte's AI cloud service brand); the "ark" in its name means "ark".

In addition to "cloud lobster farming", lobsters can also be installed on local computers.

Friends who are not familiar may be curious about the difference between the two; let me briefly explain.

First, you should know that OpenClaw is an automation software. Its purpose is to allow users to describe their needs in natural language, and it finds methods to meet these needs through a large model, then completes the tasks automatically.

When installed on a local computer (your laptop), it conveniently automates operations on local files and devices, such as "find photos taken on this day last year" or "turn off the smart light in the living room and check the electricity consumption for the past week".

When installed in the cloud, it can interact with various online services 24/7, such as "when receiving an email, automatically generate a 30-character content summary and send a notification to the phone".

So, if you need to automate operations on online services and require long-term online access or scheduled daily tasks, then "cloud lobster farming" is suitable for you.

4. Basic Operations of ArkClaw

4.1 Interface

Let me show you what ArkClaw looks like.

Enter the console, click "Create Now", and create a lobster instance.

Once created, it’s already installed and ready for use.

The interface is very simple, just a dialogue box. ArkClaw has customized the official control console for lobsters to simplify the operational interface.

4.2 Information Gathering

You can have a dialogue with the AI model in the dialogue box, which is no different from using other models.

For example, we can ask it to gather information.

You can see that since the gathered content is dynamic, the model thought of many implementation plans and finally succeeded.

Everyone should remember that ArkClaw is just a remote host; any technical solution that can be used by a server can be utilized, which is more powerful than lobsters installed on a regular personal computer.

4.3 Sending Messages

After obtaining information, lobsters can send this information to your phone.

Currently, ArkClaw supports binding with WeChat Work, DingTalk, and Feishu. Among these, Feishu, being their own product, has the simplest binding operation and is quick and convenient; just scan the code. The other two require more complicated operations, as detailed in the official documentation.

Click the "Feishu Pairing" button at the top of the dialogue box. (The "Message Channel" button in front is used to bind WeChat Work and DingTalk.)

The system will open a terminal window and output a QR code. After scanning with Feishu, you can create a bot that binds with the current ArkClaw instance.

Through this bot, you can dialogue with the current ArkClaw instance on your phone.

You can also send messages to your phone from the ArkClaw web console on your computer.

After entering the command above on the computer, your phone will receive the message (see the image below).

4.4 Scheduled Tasks

We can also specify the time and frequency for lobsters to execute certain tasks, which are called scheduled tasks.

First, use natural language to set scheduled tasks in the dialogue box.

Once set, your phone will receive messages daily.

If you want to delete a scheduled task, you can also issue a command in natural language.

5. Skills and Other Settings

5.1 Skills

The capabilities of lobsters themselves are limited, and they will encounter some problems that they do not know how to handle. At this time, you can expand their capabilities through Skills, which greatly increases the utility of lobsters.

What are Skills? Simply put, they are...

This records the technology content worth sharing every week, released on Fridays.

This magazine is open source, and submissions are welcome. There is also a service called "Who is Hiring" that publishes programmer job information. For collaboration, please contact via email (yifeng.ruan@gmail.com).

Cover image

A scenic area in Fuling, Chongqing has set up the world's first "Giant Stone Suspension Bridge," where the bridge deck consists of large stones, and one could easily step into thin air if not careful. (via)

Testing is the new moat

Next.js is currently the number one JS framework. In my estimation, half of the JS full-stack applications I encounter are developed using it.

Two weeks ago, this framework was upended by a piece of news.

A Cloudflare engineer announced that he only used one week to re-implement Next.js with AI, naming it vinext.

In fact, the product prototype was generated in just one day, with the following days spent on refinement.

"The actual work began on February 13. By that evening, the basic functionality was already implemented. By the afternoon of the next day, 10 out of 11 routers were completed. On the third day, it was already deployed to our server, achieving complete client hydration.

In the next few days, the focus was mainly on security hardening: fixing edge cases, expanding the test suite, and increasing API coverage to 94%."

This new implementation performs better than the original Next.js.

"In early benchmark tests, the build speed improved fourfold, and the size of the client package shrank by 57%. Production environment Next.js applications are already running directly on it."

The code for this vinext has been released.

I think this event is a significant blow to Next.js.

Next.js is a product of Vercel, backed by a large development team, with substantial investment each year, having been developed for a full 10 years. Although it is open-source software, the enterprise edition, cloud services, plugins, and skins all come at a cost, with last year's annual revenue reaching 200 million dollars.

This seemingly insurmountable moat is vulnerable in the face of AI. An engineer replicated the work of a large team over ten years in just one week, and existing web applications can run without changing a single line of code, supporting all functionalities of the original version.

Do you know how much it cost? The token fee was only 1100 dollars!

How can Vercel justify further investment in the development of Next.js, and how can clients be willing to pay high usage fees for specific features?

Broadly speaking, all commercial software has been heavily impacted. The code moat no longer exists; with a small investment, AI can replicate large software.

So, to protect themselves, software companies will definitely take steps to prevent AI replication next.

How to defend? The key is test cases.

The reason the Cloudflare engineer was able to successfully replicate is primarily that Next.js has comprehensive documentation, a vast community of articles, and complete test cases. Any API simulated by AI, as long as it can pass the original interface tests, can be confirmed to be 100% compatible.

If test cases cannot be obtained, who knows if the code behaves consistently? Who would dare to run it in a production environment?

It can be imagined that to prevent replication, large software projects will certainly protect their test cases. Testing is the new moat.

The world's most popular database, SQLite, has 156,000 lines of code, but its test cases amount to 92.05 million lines, a staggering 590 times larger!

Among them, the core test suite TH3 is closed-source and not publicly available, mainly testing extreme conditions and edge cases in critical industries like aviation and healthcare, considered a core technological asset. It is these confidential test cases that make SQLite difficult to replicate.

Coincidentally, just a couple of days ago, another open-source project, tldraw, also prepared to close-source its test cases.

To be honest, keeping test cases confidential is certainly not conducive to the development of open-source projects, but developers need to protect their interests. In the face of increasingly powerful AI, more and more software may choose to do this.

AI replication copyright issues

There is also a copyright issue with AI replication software, which has caused significant controversy.

Next.js is under the most permissive MIT license, so replication does not have copyright issues. However, someone replicated a project called chardet, which sparked a huge controversy.

Chardet originally used a more restrictive LGPL license, but after replication, it was changed to the MIT license, leading to protests from the original author.

Opinions online are divided into two camps.

Supporters argue that AI only replicated the functionality and interface, and the code is completely different, so it can change the license.

Opponents argue that the GPL stipulates that all derivative works cannot change their license, and AI replication falls under this category.

What’s more troublesome is that U.S. law states that AI-generated products have no copyright and belong to the public domain. This means that AI-replicated software cannot set licenses, and any such setting would be invalid.

According to this law, software licenses become quite meaningless. Regardless of what license it is, anyone can replicate it with AI and circumvent it; AI-implemented versions have no copyright at all.

Technology dynamics

1. AI rewrites profanity

The gaming platform Roblox announced that it will use AI to modify player dialogues in real-time to make them more civilized.

Previously, if players cursed in the game, the system would only filter it, displaying as ####, but you could still tell they were cursing.

Now, AI will completely rewrite the entire sentence to make the expression more polite and civilized, so you won't notice the other person is cursing.

Although this might seem a bit false, it is indeed necessary. Online forums should follow suit to avoid personal attacks ruining the atmosphere of communication.

2. Laser Internet on planes

The European Space Agency successfully conducted an experiment on "laser internet" for planes, through

1. Introduction

In the past two days, has everyone been bombarded by "lobster" (OpenClaw)?

There are news articles about it everywhere, even the representatives at the Two Sessions and Xinhua News Agency are discussing it. It’s truly astonishing that an AI software can create such a big stir.

People are enthusiastic, with free offline installation events packed, and the online "paid installation" business thriving.

Many might still be unaware that there is now the simplest way to use lobster: ArkClaw.

So simple that you don’t need to worry about installation at all, because it is a no-install solution, directly built-in with lobster, ready to use out of the box.

I just started using it yesterday and can’t wait to share my initial impressions. For those who haven’t tried it, you can consider it as a tutorial titled "Zero Threshold Getting Started with Lobster" to see what lobster is all about.

2. What is ArkClaw

Here’s the thing, old readers might remember that I evaluated Byte's latest Seed 2.0 model before the Spring Festival.

In the article, I mentioned that this is Byte's strongest base model so far, used by the mobile Doubao, and its testing performance is very good.

Later, the colleagues at Byte gifted me the Coding Plan package to facilitate further testing of this model, allowing various AI programming tools to call its API (of course, the package also includes other domestic models for free use).

This Monday, I suddenly discovered that Byte's Coding Plan package has launched a bundled service, which is ArkClaw.

I asked customer service and learned that as long as you activate the Coding Plan now, you can use lobster for free.

In other words, as long as you use Byte's AI programming package, without spending an extra penny, Byte provides a remote host with lobster already installed, which you can use freely.

It should be noted that the Coding Plan is divided into two packages: lite (first month 9.9 yuan) and Pro (first month 49.9 yuan). The lite package can only be experienced for free for 7 days, while only the Pro package can use ArkClaw long-term.

3. Cloud Lobster Farming

ArkClaw belongs to "cloud lobster farming" (also known as "cloud lobster"), which means installing lobster (OpenClaw) on a cloud host of Volcano Ark (Byte's AI cloud service brand), where the ark in its name means "ark".

Besides "cloud lobster farming", you can also install lobster on local computers.

Friends who are unfamiliar might wonder about the difference between the two; let me briefly explain.

First of all, you should know that OpenClaw is an automation software, and its function is to allow users to describe requirements in natural language. It finds methods that meet the needs through a large model and then automatically completes them.

When installed on a local computer (your laptop), it conveniently automates local files and devices, for example, "find photos taken on this day last year" or "turn off the smart light in the living room and check the electricity consumption for the last week".

When installed in the cloud, it can interact with various online services 24/7, for instance, "when an email is received, automatically generate a 30-character content summary and send a notification to the phone".

Therefore, if you need to automate operations on online services and require long-term online presence or scheduled daily runs, then it is suitable to use "cloud lobster farming".

4. Basic Operations of ArkClaw

4.1 Interface

Let me show you what ArkClaw looks like.

Enter the console, click "Create Now", and create a lobster instance.

Once created, it is already installed and can be used directly.

The interface is very simple, just a dialog box. ArkClaw has customized the official console for lobster, simplifying the operation interface.

4.2 Information Retrieval

You can interact with the AI model in the dialog box, which is no different from the use of other models.

For example, we can ask it to retrieve information.

As you can see, since the content is dynamic, the model thought of many implementation plans and successfully completed it in the end.

Remember, ArkClaw is just a remote host; any technical solutions available on any server can be used, which makes it more powerful than lobster installed on a typical personal work computer.

4.3 Sending Messages

After obtaining information, lobster can send this information to your phone.

Currently, ArkClaw supports binding with WeChat Work, DingTalk, and Feishu. Among them, Feishu, being their own product, has the simplest binding operation, which is quick and convenient, just scan the code. The other two are more complicated; refer to the official documentation for details.

Click the "Feishu Pairing" button at the top of the dialog box. (The previous "Message Channel" button is for binding WeChat Work and DingTalk.)

The system will open a terminal window, output a QR code, and after scanning with Feishu, you can create a robot that binds to the current ArkClaw instance.

With this robot, you can chat with the current ArkClaw instance on your phone.

You can also send messages to your phone through the ArkClaw web console on your computer.

After entering the above command on the computer side, the phone side will receive a push notification (as shown below).

4.4 Scheduled Tasks

We can also specify the time and frequency for lobster to execute certain tasks, which is called a scheduled task.

First, use natural language to set up the scheduled task in the dialog box.

Once set up, your phone will receive messages every day.

If you want to delete the scheduled task, you can also issue the command in natural language.

5. Skills and Other Settings

5.1 Skills

Lobster's own capabilities are limited, and there will always be some problems that it doesn't know how to handle. At this point, you can expand its capabilities through skills, greatly increasing the use cases for lobster.

What is a skill? Simply put...

This records the technology content worth sharing every week, published on Fridays.

This magazine is open source and welcomes submissions. There is also a service called "Who is Hiring", which publishes programmer job information. For cooperation, please contact via email (yifeng.ruan@gmail.com).

Cover Image

A scenic spot in Fuling, Chongqing, has built the world's first "Giant Stone Suspension Bridge", where the bridge surface consists of large stones, and one might accidentally step into the void. (via)

Testing is the new moat

Next.js is currently the number one JS framework. I estimate that half of the JS full-stack applications I encounter daily are developed using it.

Two weeks ago, this framework was overturned by a piece of news.

A Cloudflare engineer announced that he had used AI to reimplement Next.js in just one week, naming it vinext.

In fact, a product prototype was generated in one day, and the following days were just for refinement.

"The actual hands-on work started on February 13, and by that evening, basic functionality had been achieved. The next afternoon, 10 out of 11 routers were completed. By the third day, it had been deployed to our servers, achieving complete client hydration.

In the following days, the focus was mainly on security hardening: fixing extreme cases, expanding the test suite, and increasing API coverage to 94%."

This new implementation outperforms the original Next.js.

"In early benchmarks, build speed increased fourfold, the size of client packages decreased by 57%, and Next.js applications in production environments are already running directly on it."

The code for vinext has already been released.

I believe this is a significant blow to Next.js.

Next.js is a product of Vercel, backed by a large development team that has made substantial investments every year for a full decade. Although it is open-source software, the enterprise version, cloud services, plugins, and skins are all charged, with last year's annual revenue reaching $200 million.

This seemingly insurmountable moat is no match for AI. An engineer replicated the results of a large team’s ten years of work in a week, with existing web applications able to run without changing a line of code, supporting every feature of the original version.

Do you know how much it cost? The token cost was only $1,100!

How can Vercel justify further investment in Next.js development, and how can clients be willing to pay high usage fees for certain features again?

Broadly speaking, all commercial software has suffered a heavy blow. The moat of code no longer exists; with a small investment, AI can replicate large software.

So, to protect themselves, software companies must take steps to prevent AI replication next.

How to prevent it? The key is test cases.

The reason the Cloudflare engineer was able to successfully replicate was mainly due to Next.js having complete documentation, a vast community of articles, and comprehensive test cases. Every API simulated by AI can be confirmed to be 100% compatible as long as it passes the existing interface tests.

If test cases cannot be accessed, who knows whether the code behavior is consistent? Who dares to run it in a production environment?

It is easy to imagine that to prevent replication, large software projects will definitely protect their test cases. Testing is the new moat.

The world's most popular database SQLite has 156,000 lines of code, but 92,050,000 lines of test cases, a staggering 590 times larger!

Among them, the core test suite TH3 is closed source and not public, mainly testing extreme situations and edge cases in critical industries like aviation and healthcare, constituting core technological assets. It is precisely these confidential test cases that make SQLite difficult to replicate.

Coincidentally, just a couple of days ago, another open-source project tldraw also prepared to close-source its test cases.

To be honest, confidential test cases are certainly not conducive to the development of open-source projects, but developers need to protect their own interests. In the face of increasingly powerful AI, more and more software may choose to do the same.

AI Replication Copyright Issues

AI replication of software also brings copyright issues, which have sparked significant controversy.

Next.js has the most permissive MIT license, so replication poses no copyright issues. However, someone replicated a project called chardet, which has caused a huge controversy.

Chardet originally used a more restricted LGPL license, but after replication, it was changed to an MIT license, provoking protests from the original author.

Opinions online are divided into two camps.

Supporters argue that AI simply replicated the functions and interfaces, the code is completely different, so it can change the license.

Opponents say that the GPL stipulates that all derivative works cannot change the license, and AI replication belongs to derivatives.

More troubling is that U.S. law states that AI-generated products have no copyright and are considered public domain. This means that AI-replicated software cannot set a license; any license set would be invalid.

According to this law, software licenses become largely meaningless. No matter what license you have, anyone can replicate via AI, and versions created by AI have no copyright whatsoever.

Technology Dynamics

1. AI Rewrites Profanity

The gaming platform Roblox announced that it will use AI to modify player dialogue in real-time to make it more civilized.

Previously, if players swore in the game, the system would only filter it and display it as ####, and you would still know they were cursing.

Now, AI will completely rewrite the entire sentence, making the expression more polite and civilized, so you won't even notice the other person is cursing.

Although this may seem somewhat false, it is indeed necessary. Online forums should also follow suit to prevent personal attacks from ruining the communication atmosphere.

2. Laser Internet for Airplanes

The European Space Agency successfully conducted an experiment on "laser internet" for airplanes, through

1. Introduction

In the past two days, has everyone been bombarded with news about "Lobster" (OpenClaw)?

Its news is everywhere, even the representatives at the Two Sessions and Xinhua News Agency are discussing it. It’s truly astonishing that an AI software can cause such a huge reaction.

People's enthusiasm is high, with crowded free offline installation events, and online "paid installation" businesses thriving.

Many people probably still don't know that there is now the simplest way to use Lobster: ArkClaw.

So simple that you don't need to worry about installation at all, because it's a no-installation solution that comes with Lobster built-in, ready to use right out of the box.

I only started using it yesterday and can't wait to share my initial impressions with everyone. For those who haven't used it yet, you can treat this as a tutorial on "Getting Started with Lobster with Zero Barriers" to see what Lobster is all about.

2. What is ArkClaw

Here's the thing: old readers may remember that I reviewed Byte's latest Seed 2.0 model before the Spring Festival.

In the article, I mentioned that this is Byte's strongest base model so far, and the mobile Doubao uses it, with great testing performance.

Later, Byte's colleagues gifted me a Coding Plan package for continued testing of this model, allowing various AI programming tools to call its API (of course, the package also includes other domestic models, which are free to use).

This Monday, I suddenly discovered that Byte's Coding Plan package has activated a bundled service, which is ArkClaw.

I asked customer service and learned that as long as you activate the Coding Plan now, you can use Lobster for free.

In other words, as long as you use Byte's AI programming package, you don't have to spend an extra penny, and Byte provides a remote host that has Lobster installed, which you can use freely.

It should be noted that the Coding Plan is divided into two packages: lite (first month 9.9 yuan) and Pro (first month 49.9 yuan). The lite package can only be experienced for free for 7 days, and only the Pro package allows long-term use of ArkClaw.

3. Cloud Lobster

ArkClaw belongs to "Cloud Lobster" (also known as "Cloud OpenClaw"), which means installing Lobster (OpenClaw) on the cloud host of Volcano Ark (Byte's AI cloud service brand), where the word ark means "ark".

Besides "Cloud Lobster", you can also install Lobster on a local computer.

Friends who are not familiar may wonder what the difference is; let me briefly explain.

First, you should know that OpenClaw is automation software, which allows users to describe needs in natural language. It finds methods to meet the needs through a large model and then completes them automatically.

When it is installed on a local computer (your laptop), it conveniently automates operations on local files and devices, such as "find photos taken on this day last year" or "turn off the smart light in the living room and check the electricity consumption for the past week".

When it is installed in the cloud, it can interact with various online services 24/7, such as "generate a 30-character content summary automatically when receiving an email and send a notification to the phone".

So, if you need to automate operations for online services and require long-term online access or scheduled daily runs, then using "Cloud Lobster" is suitable.

4. Basic Operations of ArkClaw

4.1 Interface

Let me show you what ArkClaw looks like.

Enter the console, click "Create Now" to create a Lobster instance.

Once created, it's already installed and ready to use.

The interface is very simple, just a dialog box. ArkClaw has customized the official console for Lobster, simplifying the operation interface.

4.2 Information Retrieval

You can chat with the AI model in the dialog box, which is no different from using other models.

For example, we can ask it to retrieve information.

As you can see, since the retrieved content is dynamic, the model came up with many implementation plans and successfully completed it in the end.

Everyone should remember that ArkClaw is just a remote host; any technical solution that can be used on a server can be used with it, making it more powerful than Lobster installed on a regular personal work computer.

4.3 Sending Messages

After obtaining information, Lobster can send this information to your phone.

Currently, ArkClaw supports binding with WeChat Work, DingTalk, and Feishu. Among them, binding with Feishu is the easiest since it's their own product; the operation is simple, quick, and can be done via QR code. The other two are relatively cumbersome; refer to the official documentation for details.

Click the "Feishu Pairing" button at the top of the dialog box. (The previous "Message Channel" button is used for binding with WeChat Work and DingTalk.)

The system will open a terminal window displaying a QR code, which can be scanned by Feishu to create a bot that links to the current ArkClaw instance.

Through this bot, you can now chat with the current ArkClaw instance on your phone.

You can also send messages to your phone from the ArkClaw web console on your computer.

After entering the command on the computer, the phone will receive a message (as shown in the picture below).

4.4 Scheduled Tasks

We can also specify the time and frequency for Lobster to perform certain tasks, which are scheduled tasks.

First, use natural language to set up scheduled tasks in the dialog box.

Once set up, your phone will receive messages every day.

If you want to delete a scheduled task, you can also issue commands in natural language.

5. Skills and Other Settings

5.1 Skills

Lobster's abilities are limited and will always encounter some problems it doesn't know how to handle. In these cases, you can expand its capabilities through Skills, greatly increasing the usefulness of Lobster.

What is a Skill? Simply put, it is a way to enhance Lobster's functionality.

This records the tech content worth sharing every week, released on Fridays.

This magazine is open source, and contributions are welcome. There is also a service called "Who Is Hiring" that publishes job listings for programmers. For collaboration, please contact via email (yifeng.ruan@gmail.com).

Cover Image

A scenic area in Fuling, Chongqing has set up the world's first "Giant Stone Suspension Bridge," where the bridge deck consists of large stones, and one might accidentally step into thin air. (via)

Testing is the new moat

Next.js is currently the top-ranked JS framework. I estimate that half of the JS full-stack applications I've encountered are developed using it.

Two weeks ago, this framework was overturned by a piece of news.

A Cloudflare engineer announced that he replicated Next.js using AI in just one week, naming it vinext.

In fact, a product prototype was generated in a day, and the following days were just for refinement.

"The real work began on February 13. By that night, the basic functionality was implemented. The next afternoon, 10 out of 11 routers were completed. By the third day, it was already deployed on our servers, achieving full client hydration.

In the following days, the focus was mainly on security hardening: fixing edge cases, expanding the test suite, and increasing API coverage to 94%."

This new implementation outperforms the original Next.js.

"In early benchmark tests, build speed improved by 4 times, the size of client packages shrank by 57%, and production Next.js applications are already running directly on it."

The code for this vinext has already been released.

I believe this is a significant blow to Next.js.

Next.js is a product of Vercel, backed by a large development team, with huge investments every year, having been developed for a full 10 years. Although it is open-source software, the enterprise version, cloud services, plugins, and skins are all charged. Last year's revenue reached $200 million.

This seemingly insurmountable moat crumbles before AI. An engineer replicated the results of a large team’s decade-long work in just one week, and existing web applications can run without changing a line of code, fully supporting every feature of the original.

Do you know how much it cost? The token fee was merely $1,100!

How can Vercel invest more money into the development of Next.js, and how can clients be willing to pay high usage fees for a feature?

By extension, all commercial software has been severely impacted. The code moat no longer exists; with just a small financial investment, AI can replicate large software.

So, to protect themselves, software companies will definitely take steps to prevent AI replication.

How to defend against it? The key is test cases.

The reason the Cloudflare engineer was able to replicate successfully this time is primarily that Next.js has comprehensive documentation, a vast community of articles, and complete test cases. Every API simulated by AI can be confirmed to be 100% compatible as long as it passes the existing interface tests.

If test cases cannot be obtained, who knows if the code behavior is consistent, and who dares to run it in a production environment?

It can be imagined that to prevent replication, large software projects will definitely protect their test cases. Testing is the new moat.

The world's most popular database, SQLite, has 156,000 lines of code, but 92,050,000 lines of test cases, a staggering 590 times larger!

Among them, the core test suite TH3 is closed source and not made public, mainly testing extreme situations and edge cases in critical industries such as aviation and healthcare, which belong to core technical assets. It is precisely these confidential cases that make it difficult to replicate SQLite.

Coincidentally, just a few days ago, another open-source project, tldraw, also prepared to close source its test cases.

To be honest, keeping test cases confidential is certainly not conducive to the development of open-source projects, but developers need to protect their own interests. In the face of increasingly powerful AI, more and more software may choose to do this.

AI Replication Copyright Issues

AI replication software also has a copyright issue that has caused a lot of controversies.

Next.js is under the most permissive MIT license, so there are no copyright issues with replication. However, some have replicated a project called chardet, leading to significant controversy.

Originally, chardet used a more restrictive LGPL license, but after replication, it was changed to the MIT license, prompting protests from the original author.

Opinions online are also divided into two factions.

Supporters say that AI only replicated the functionality and interfaces, and the code is completely different, so it can certainly change the license.

Opponents argue that the GPL stipulates that all derivative works cannot change the license, and AI replication belongs to derivative.

What's more troublesome is that U.S. law states that AI-generated products have no copyright and are in the public domain. This means that AI-replicated software cannot set licenses, and any set license is invalid.

According to this law, software licenses become less meaningful. Regardless of what license you have, anyone can replicate it with AI, and the versions implemented by AI have no copyright.

Tech Dynamics

1. AI Rewrites Profanity

The gaming platform Roblox announced that it will use AI to modify players' dialogues in real time to make them more civilized.

Previously, if players swore in the game, the system would only filter it, displaying as ####, and you would still know they were cursing.

Now, AI will rewrite the entire sentence, making the expression more polite and civilized, so you won't notice the other person is swearing.

Although this seems a bit fake, it is indeed necessary. Online forums should also follow suit and not let personal attacks ruin the atmosphere of communication.

2. Laser Internet for Aircraft

The European Space Agency successfully conducted an experiment for "laser internet" on aircraft through