With the popularity of DeepSeek and Beanbag, I believe that more and more people are beginning to choose to use AI to assist in writing in their life and work, whether it is to organize the material, or AI to embellish the details, or even AI to form a text with one click ......
Efficiency has increased, but with that comes another problem: each platform is quietly performing AI detection in the background.
AI to recognize AI, although it can not achieve the recognition rate of 100%, there is a possibility of false alarms, but for articles containing high AI rate, the platform to limit the flow is also not soft.
Whether you're an AI ghostwriter who wants to measure the AI rate, or you're working on an AI We MediaWhen dealing with the platform's AI detection, it is necessary to master some 'de-AI' techniques.
So today I'm going to share some quick tips on how to deal with the two major techniques of the official Tencent Zhuzhu AI detection assistant, which I hope will be helpful to you.
address:https://www.1ai.net/34887.html
PS: Everyone must recognize this official Tencent link above.
Tip 1: Use AI to Remove AI Taste
I first put the last article on the dismantling of the top ten paths to realizing the dismantling of the intelligent body, and washed it with AI, which was detected by the Jubilee AI text, and was very sure that it was the AI concentration of 100%.
Reduce the concentration of AI, the simplest way is definitely to let the AI itself to remove the AI flavor, I tested a variety of cue words in multiple models, found that the cue words do have an impact, but the biggest impact is still the AI model itself, so in the same premise as the instruction -
DeepSeek removes the AI flavor, yet still has the AI concentration of 100%.
The beanbags look like this too, in a word removing the AI flavor is not effective.
Relatively speaking, the foreign monk can read the scriptures, and of the three players, ChatGPT, Gemini, and Gork, the best performer was Gemini, who reduced the AI concentration from 100% to around 40% in one breath.
But this is not over, I suddenly thought that also fell a player: is the Tencent family of mixed big model, that is, Yuan Bao, I did not expect this time directly end the game, a breath of AI concentration down to 25%!
This is almost a sentence to reduce the limit of AI concentration, I have tried to take the Jubilee AI text detected passages out of the AI alone to modify, but contains AI rate is often not down but up.
Because AI generation has a certain degree of randomness, each time to let the AI to reduce the AI flavor is unstable, so it is necessary for us to analyze the Tencent Yuanbao in the process of this de-AI flavor in the end did what to let the AI concentration plummeted, I summarized-
1, increased the specific case: increase the scene details and personal experience, is the core of the Yuanbao can reduce the concentration of AI, instead of letting the model to the text to prune the branches, why not add a case you observed, a breath of fresh air can make the article more real.
2. Spoken expression: this step is actually done by every AI, but Yuanbao's tone is closer to personal sharing, and the paragraph articulation is more natural.
3, weaken the sense of structure and listing: AI writing will be too neat, that is, the so-called mechanical flavor, but do not know that imperfect will be more real.
In the future, when you want AI to remove the AI flavor, why don't you start from these 3 perspectives? Can we quickly reduce the AI concentration to 0%? Yes, I'd like to share another popular tricky solution here.
Tip 2: The Great Punctuation Revision
Regular writers must know that punctuation is actually divided into Chinese full-width and English half-width.
Chinese full-corners are punctuation marks typed under Chinese input methods--
:,...? ; ""()-
English half-corners are also known as punctuation marks typed under the English input method -
:,...?! (""()-
Generally speaking, Chinese articles will use Chinese full-corner, and English articles will use English half-corner, and the AI big model does the same in the output.
What about if it is reversed? Chinese articles with English half-corners, this kind of error will be considered by the AI detection assistant as a human error, and the AI concentration will be instantly reduced.
Articles like the one where metabolism was lowered to 25% AI concentration were revised to be lowered to 0%.
Even the original 100% AI concentration was reduced to 0% when so manipulated, that's how exaggerated it was.
Instead of modifying punctuation one by one, you can have any of the big AI models help us generate a web tool for symbol conversion with a simple prompt word:
Make a public punctuation conversion online web tool to lower the Chinese full-width punctuation in the article and modify it to English half-width punctuation.
The full-angle symbols are,.? :; ""()-
The semicolon symbols are,...?! :;""()-
Save it locally and you can use it again and again:
However, we should know that this is a loophole for drilling AI detection, which also sacrifices a certain amount of readability, and is very easy to fix officially, and is not suitable for long term application.
In fact, according to the conclusion of the editorial observation: when the AI concentration of an article is reduced to below 30%, the article will not appear to limit the flow of the cliff, the real decision of the article flow is still the selection of the title - title - cover three major basic content, which affects the article can cater to the algorithm to obtain the core of the recommendation.
Well, that's all for today, believe in yourself, stick with it, and hopefully today's content will help you.