AI Writing Markers: Building an Open Dataset and Checker on Hugging Face

Описание и применение

Инструмент ai‑writing‑markers предоставляет прозрачный, источник‑подтверждённый анализ текста, выявляя паттерны, характерные для LLM‑выхода, и позволяет авторам редактировать работу без зависимости от непрозрачных вероятностных оценок.

Задача

Commercial AI detectors give opaque probability scores with high false‑positive rates, especially against non‑native English writers.

Результат

The checker provides transparent, source‑backed feedback on AI‑like writing patterns, empowering users to edit drafts without relying on opaque probability scores.

Рабочий процесс

• Clone the ai‑writing‑markers repo
• Run python3 check.py <file> or pipe text to the checker
• Review the per‑category hit counts and source‑backed markers
• Edit the draft to reduce dense clusters of flagged patterns
• Re‑run the checker to confirm improvements

Доказательства

Markers are weak signals. One instance of "delve" means almost nothing.

The honest thing a tool can do is point at the patterns that make writing read as machine‑generated — so a human can decide whether to change them.

The checker loads markers.json, runs lexical matching per category, applies regex patterns for structural and formatting tells, and computes burstiness plus type‑token ratio as perplexity proxies.