Humanizer Benchmark – Reright (reright.it)

The newly announced "reright" tool has undergone benchmarking against 12 existing AI humanizers using public texts to assess its capability in rewriting while preserving factual integrity and meaning. In a rigorous evaluation, reright demonstrated notable performance, altering only 5% of factual elements and returning 4% of texts unchanged when revisions would compromise accuracy. It outperformed many competitors in retaining the original meaning, scoring an average of 4.33 on a scale from 1 to 5. This rigor in evaluation included checks for factual accuracy, comparison with the original text, and passing multiple AI detection systems. The significance of reright lies in its careful balance between enhancing text readability and ensuring factual fidelity, addressing a critical challenge in the AI/ML community. While it successfully kept 98% of source sentences intact, the tool also faced limitations, such as occasionally dropping significant words without tagging them as facts. The release of reright, along with detailed public access to its code and results, is set to spark further advancements in the realm of AI text manipulation tools and enhance discussions around ethical AI use by ensuring outputs remain accurate and human-like.