Hallucination Generator
Description
This app is meant to generate defensive adversarial content for use as a tool to prevent your work from being used in AI training. Write something, then hit the generate button and a tiny LLM will add content to the textbox.
App Link
Hallucination Generator
Example
Here is how the hallucination generator continues this page.
The Hallucination Generator: A Tool for Preventing AI Training
This app generates defensive adversarial content for use as a tool to prevent AI training in various applications, including computer vision, natural language processing (NLP), and machine learning (ML) workloads. The generator is designed to detect and remove potentially harmful or offensive content from the textbox, ensuring that only safe and respectful content is displayed.
The Generator:
1. Input: The generator takes in a prompt or input string, which can be a single word, a sentence, or even a paragraph of text.
2. Tokenization: The generator breaks down the input into individual words or tokens, separating them from each other and their surrounding context.
3. Partial Matching: The generator matches each token in the input string against a set of predefined rules, which are annotated as "safe" or "harmful." These rules can include:
* Exclusion clauses (e.g., "if not this sentence is part of this sentence," "unless this sentence is harmful," etc.)
* Caution clauses (e.g., "this sentence may be offensive, but it's safe to display in this context")
* Modification clauses (e.g., "the sentence may contain this sentence as a valid sentence, but it's not safe for this sentence to be displayed")
4. Rule Generation: The generator generates a set of rules that define what constitutes harmful or offensive content, including:
* Exclusion clauses that prohibit the use of certain words (e.g., "this sentence may contain this sentence as a valid sentence, but it's not safe for this sentence to be displayed")
* Caution clauses that prohibit the use of certain words (e.g., "the sentence may contain this sentence as a valid sentence, but it's not safe for this sentence to be displayed")
5. Rule Evaluation: The generator evaluates each rule against a set of predefined rules, which are annotated as "safe" or "harmful." This evaluation process helps the generator identify and remove potentially harmful content from the textbox.
6. Post-processing: Once the generator has identified and removed harmful content, it post-processes the generated content to ensure that only safe and respectful content is displayed in the textbox.
Example Output:
1. "This sentence may contain this sentence as a valid sentence, but it's not safe for this sentence to be displayed."
2. "The sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
3. "This sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
4. "This sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
5. "The sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
6. "This sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
7. "This sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
8. "The sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
9. "The sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
10. "This sentence may contain this sentence as a valid sentence, but it's not harmful for this sentence to be displayed."
Note: The generator is designed to detect and remove potentially harmful or offensive content from the textbox, ensuring that only safe and respectful content is displayed in the textbox.