How is Ai content determined?

So I see posts here of platforms banning AI work. I am not the most tech savvy. How does this work? Is there programming data underneath that identifies it as AI versus human work? Or is it merely the style/ lack of imperfections.

If a human writes perfectly in AI style writing, can it be confused?

If an ai writes something and you guys edit it…. Take what it says and alter it to lower the plagiarism score does that have an effect?

It is determined by another GPT model that does the same prediction on tokens and compares the results. It's a very simple and quite accurate technique. However, it can be easily tricked by a simple synonymizer made 20-30 years ago...

you are right if the idea of AI is to get content similar or bettr than human writing why is easy to detect it, some AI detectors just count the ammount of common, uncommon and rare words to determine if is human or no, and not all AI detectors work the same or give the same result over a text generated by AI to me AI detectors don't work, something else that could determine AI content is if you copy and paste the result, if you ask something to gpt question 1 and someone else in a different computer ask the same question both will get the same response unless you use a prompt and gpt do not modify change anything unless you ask for it and the information that collect is from the internet so can get you copyright or no copyright information by copy and paste segments is posible for you to find the source at the end AI detectors are using AI too sometimes some are real fake AI detectors.

MoonStarz said:

you are right if the idea of AI is to get content similar or bettr than human writing why is easy to detect it, some AI detectors just count the ammount of common, uncommon and rare words to determine if is human or no

Click to expand...

No, they work differently. I explained this earlier.

I asked the AI itself :

AI-generated content can be determined through various methods depending on the specific task or application. Here are some common approaches:

1. Training on Data: Many AI models, particularly in natural language processing (NLP), are trained on large datasets of text. These datasets can include books, articles, websites, social media posts, and more. The model learns patterns, grammar, and semantics from this data, which it then uses to generate new content.
2. Fine-Tuning: Some AI models are pre-trained on generic datasets and then fine-tuned on specific datasets relevant to a particular task or domain. For example, a model pre-trained on general English text might be fine-tuned on medical literature to generate medical content.
3. Prompt-Based Generation: In this approach, users provide a prompt or a starting point for the AI to generate content. The AI then uses the prompt to guide its generation process. The quality and relevance of the generated content often depend on the specificity and clarity of the prompt.
4. Objective Functions: In reinforcement learning setups, AI models are trained to maximize or minimize a certain objective function. For content generation, this could mean maximizing readability, coherence, relevance, or other metrics. The model generates content iteratively, receiving feedback on its performance from the objective function, and adjusts its approach accordingly.
5. Conditional Generation: AI models can be conditioned on various factors such as keywords, topics, or stylistic preferences. For instance, a model might be conditioned to generate content in the style of a particular author or to focus on a specific aspect of a topic.
6. Human Feedback: Some AI content generation systems incorporate human feedback to improve the quality of generated content. Users can provide feedback on the generated content, which is then used to retrain the model or fine-tune its parameters.

Google bans your sites for poor user experience, not for positive AI detection results. Google doesn't care about the AI or non-AI origin of your content. It only cares about its quality. You may not know this, but today Google doesn't even need algorithmic methods to check the quality of content. It has a billion army of human bots that do this job for free every day. Every time you open a site in Chrome, every time you open a page with embedded Google Analytics code, every time you browse the web with Android, you are working for Google. Remember that simple fact.

The ban does not necessarily imply that they are AI works, it may be an abuse of AI. I saw a post here where someone blindly copied 'As of my latest update in January 2022' into an article uploaded on Google. Please tell me how that should be handled.

flackz said:

Google bans your sites for poor user experience, not for positive AI detection results. Google doesn't care about the AI or non-AI origin of your content. It only cares about its quality. You may not know this, but today Google doesn't even need algorithmic methods to check the quality of content. It has a billion army of human bots that do this job for free every day. Every time you open a site in Chrome, every time you open a page with embedded Google Analytics code, every time you browse the web with Android, you are working for Google. Remember that simple fact.

Click to expand...

true. also google doesn’t spend billions of dollars to check each and every page for AI generated content as far as I know (since it’s almost the same resource intensive task as to generate that content at the first place), all that just doesn’t fit their business model and has no fucking sense at all.

flackz said:

Google bans your sites for poor user experience, not for positive AI detection results. Google doesn't care about the AI or non-AI origin of your content. It only cares about its quality. You may not know this, but today Google doesn't even need algorithmic methods to check the quality of content. It has a billion army of human bots that do this job for free every day. Every time you open a site in Chrome, every time you open a page with embedded Google Analytics code, every time you browse the web with Android, you are working for Google. Remember that simple fact.

Click to expand...

Correct. Most high authority sites now use AI and google could careless because they provide really value. Edited and fine tined to answer user queries.

Try improve your content quality and google judge quality not AI detection or something

If you are conversant with AI-generated content, you don't need a detector to know the tone.

Platforms often use patterns and language analysis to detect AI-generated content. While a human writing perfectly in AI style might confuse detection, it's best to aim for originality. Editing AI-generated content to lower plagiarism scores might help, but it's still risky.