Nonio

Nonio is a sophisticated measuring instrument designed to analyze text without making claims of authorship. This tool estimates statistical signals of machine-generated language and reports findings with associated uncertainty. Importantly, it does not assert whether a text is written by an AI or a human; rather, it provides valuable metrics to aid in understanding the nature of the content.

Key Features

• Precision Measurement: Just like a sliding vernier scale enhances measurement accuracy, Nonio provides additional precision in evaluating texts, reporting statistical signals along with false-positive rates segmented by text category.

• Unbiased Reporting: Nonio issues results without making judgments. For instance, it does not output that a text is "87% AI", which has led to significant misinterpretations in academic and professional contexts.

• Local Operation: The tool runs entirely locally, ensuring data privacy and reducing the risk of unintentional data exposure. Calibrated primarily for Spanish texts, it commits to reliable outputs while protecting users from erroneous classifications.

How It Works

Nonio employs two independent signal families for analysis:

1. Fast-DetectGPT: This signal focuses on the conditional probability curvature, making it computationally efficient and capable of running on standard CPUs.
2. Binoculars: This signal measures perplexity ratios between different models to evaluate the intrinsic difficulty of the text.

This dual approach ensures variability in signal interpretation, requiring consensus between the two to generate a report. If they disagree significantly, Nonio abstains from providing a measurement rather than risking misinformation.

Usage Example

Nonio can be easily operated via command line:

$ nonio analyze essay.md
⚠  Nonio does not assert authorship. This output must not be the sole basis for any academic, employment, or editorial decision.

fuente: essay.md
perfil: qwen2.5-0.5b   idioma detectado: es
RESULTADO: señal MODERADO
  umbral aplicado:     0.9167
  falsos positivos:    5.3%

Additionally, it allows for JSON output and can generate structured HTML reports. Every analysis includes the applied thresholds and measured error rates, highlighting the statistical nature of the readings.

Ethical Principles

Nonio adheres to several key principles:

• Instrument, Not Judge: Outputs never conclude authorship; they provide statistical insight only.
• Abstention as Validity: If evidence is insufficient, Nonio will abstain from making a claim, presenting this as a legitimate result.
• Transparency: All outputs are traceable, enhancing accountability.

In summary, Nonio stands out as a reliable, ethically guided tool that emphasizes measured signals over conclusive judgments, making it a vital instrument for appropriately assessing machine-generated text. For further details, consult the documentation to explore its features and functions.