How to distinguish ai written articles? Teach you to use the Tencent Jubilee large model detection tool "fake" AI small essay!

"The article you stayed up late to write may have been determined by the system to be an AI ghostwriter!"

Recently, the news of a college student failing his thesis because it was detected with AI-generated traces has rushed to the hot search. the proliferation of AI writing has put everyone in the academic and media circles at risk. But don't panic, Tencent has just launched the "Vermilion Bird Large ModelDetection tool", claiming to be able to recognize the AI-generated text! Today we will test - it in the end can pull out the GPT, Tongyi thousand questions these "writing masters"?

Why do we need "fake tools" in the age of AI writing?

The proliferation of AI-generated texts has led to a crisis of confidence. From student assignments to press releases, and even commercial contracts, they may be mixed with "AI essays". The birth of Tencent's Jubilee Large Model Detection Tool was designed to address this issue:

1. accurate identification: Trained by 1.4 million positive and negative samples, covering news, official documents, novels and other genres, with a detection rate of over 95%.
2. Technical Principles: Analyze the grammatical structure of the text, word patterns, invisible features (e.g., fixed sentences commonly used by AI), and even compare the "overlap" of the content generated by the larger model.
3. Application Scenario: academic checking, media review, copyright protection ...... even ordinary users can use it to check the article "AI components".

Hands-on! Use Jupiter inspection tools to "fake" major models

Step 1: Enter the testing platform

Visit the official address of Tencent (https://matrix.tencent.com/ai-detect/), no need to register, directly paste the text can be detected, limited to 20 times a day (test found that with privacy mode can be unlimited woolgathering 😂).

Step 2: Multi-model text generation

We used GPT-4, Tongyiqianqian, and deepseek to generate three paragraphs of "Spring Essay" text, and submitted them to Jubilee for testing:

• GPT-4 Generated Text: Detection result: "Suspected AI-generated content accounts for 100%"! The system prompts "AI flavor is a bit heavy".

• Thousand Questions on Tongyi: 100%, labeled "A little too AI-flavored".

• deepseek:I gave deepseek different instructions given the total collapse of the front car.

deepseek after deep thinking, but the result remains:

• Texts written by humans: In order to verify the above test, I specifically found a student essay from the Internet, and the result was this: 0%.After that, I tried to splice the AI text with the human text 1:1, and the result was surprising, not 50%, or 0%!

reach a verdictJupiter: Jupiter on the mainstream large model recognition rate is very high, but after the human text splicing, AI text can also be "hidden from view".

Anti-detection strategy: how to make AI articles "more human-like"?

If you have to use AI-assisted writing, but want to pass the test, try these "anti-trope" techniques (tested and effective!). If you have to use AI-assisted writing and want to pass the test, try these "anti-cliché" techniques (proven to work!):

1. Feed it to your historical articles: Let the AI mimic your writing style with the example instruction, "Please refer to my previous article, XXX, and adjust the wording and structure of the following text".
2. Add detail and personalization: For example, inserting real cases, personal experiences, or flirtatious spiels (e.g., "The boss said this requirement is simple ......").
3. Avoiding the classic AI talking points:: Delete clichés such as "in summary" and "it is worth noting" and use more short sentences and colloquial expressions.
4. Hybrid Multi-Model Generation: Write the frame in GPT, Claude touches it up, and finally fine-tune it manually to disperse the features.

Is AI testing a "straitjacket" or an "enabler"?

The launch of the Jubilee tool marks a new stage in AI content regulation. But there is no good or evil in technology, the key is to see how to use it:

• For creators: It is a mirror that reminds us to stay original and avoid over-reliance on AI.
• to the industry: Promote transparency and a healthy ecosystem for AI-generated content (e.g., requiring AI-generated content to be labeled with a "digital watermark").
• To the average person: An additional "anti-scamming tool", such as checking to see if a ghostwriting service is using AI to fool you!

Does your article pass the Jubilee test? Go ahead and try it!