lynote-ai/ai-text-detector

A cautious, explainable AI-like text risk analyzer for local workflows and coding agents.

What it solves

It addresses the problem of overconfident and opaque AI text detectors. Instead of providing a binary "AI or Human" verdict, it acts as a cautious triage tool that provides risk estimates and explainable signals to help humans make informed decisions rather than treating a score as absolute proof.

How it works

The tool analyzes text to generate a risk score (0-100) and a verdict (such as high_ai_likelihood or insufficient_text). It uses weighted evidence signals to explain why a text is flagged and includes a short-text guardrail to prevent overstating evidence when the sample size is too small. It is available as a local CLI, a Python API, and a portable skill for repo-aware agents like Codex and Claude Code.

Who it’s for

  • Teachers needing a cautious signal before manually reviewing student submissions.
  • Editors looking to spot formulaic or templated product reviews and guest posts.
  • Trust and Safety teams wanting to prioritize suspicious long-form posts for manual review.
  • Content QA teams comparing human and AI-assisted drafts to identify generic language.

Highlights

  • Explainable Output: Provides strongest weighted evidence signals and caveats alongside the score.
  • Agent-Ready: Packaged as a skill for integration into coding agents and local workflows.
  • Conservative Design: Prioritizes avoiding false confidence over high recall, specifically refusing to make strong claims on very short texts.
  • Reproducible Evaluation: Includes scripts to evaluate performance against the public HC3 dataset.

Related

  • Dispatch
    OpenAI AI Text Classifier Release (January 2023)

    OpenAI released a public AI text classifier in January 2023 that flags likely AI‑generated English text, but it is unreliable—identifying only 26% of AI text and mislabeling 9% of human text.

  • Dispatch
    AI-written critiques help humans notice flaws

    OpenAI research demonstrates that AI-generated critiques help human evaluators identify significantly more flaws in summaries, suggesting a scalable path for human supervision of complex AI tasks.

  • Project
    khoj-ai/openpaper

    An AI-powered research workspace for annotating and understanding scientific papers with verifiable citations and cross-paper synthesis.

  • Dispatch
    OpenAI Iterated Amplification for Complex Goal Learning

    OpenAI proposes iterated amplification, an AI safety technique that enables the specification of complex, beyond-human-scale goals by decomposing tasks into simpler sub-tasks to generate training signals.

  • Dispatch
    OpenAI Prover‑Verifier Games improve legibility of language model outputs

    OpenAI announced a prover‑verifier game training method that makes strong language models generate solutions that weaker models can verify, improving both correctness and human legibility.