STE80 Humanizer Tool Simplifies AI-Generated Text for Readability

Introducing STE80 Humanizer

STE80 Humanizer is a new tool designed to refine AI-written text, making it more akin to human-authored content. The primary goal is to enhance readability and clarity for human readers, rather than to circumvent AI detection systems. This tool addresses the increasing volume of AI-generated content that requires human review and understanding.

Methodology for Text Simplification

The tool employs two main methods. First, it applies a 'humanizer' method that identifies and removes 26 specific writing patterns commonly associated with AI, as cataloged in 'blader/humanizer' based on Wikipedia's 'Signs of AI writing'. Second, it utilizes a 'Karpathy's STE80 method' to restructure sentences, aiming for a style that is '80% of the way to ASD-STE100'. ASD-STE100 is a controlled English specification originally developed for aerospace maintenance documentation, known for its stringent constraints on clear writing.

Impact on Readability

An example provided demonstrates the transformation: a complex paragraph about Kubernetes liveness probes, initially verbose and containing AI-like phrasing, is simplified into concise, direct sentences. The revised text explains the function of liveness probes, their failure mechanisms, and the implications of misconfiguration in a more straightforward manner. This simplification helps in making technical information more accessible and understandable.

Addressing the AI Output Bottleneck

The development of STE80 Humanizer aligns with observations by Andrej Karpathy, who noted that as large language models (LLMs) take on more work, humans spend more time interpreting their output. Karpathy suggested using controlled languages like ASD-STE100 to improve LLM output readability. The STE80 Humanizer directly implements this principle, aiming to make AI-generated content easier for humans to process, thereby addressing a bottleneck in workflows involving LLMs.