[

{

"type": "tip",

"title": "Beta Data Coverage",

"text": "This explorer is currently in <strong>Beta</strong> with limited coverage. The 'Science' register is restricted to English and select models. Select GPT-4.1 mini to access multilingual data in the News register."

},

{

"type": "about",

"title": "Methodology: Volume",

"text": "<strong>Volume</strong> is based on the Lexical Alignment Score (LAS), which is derived from windowed document prevalence analysis. It offers a fully automated metric for divergence, eliminating the need for manual curation. Details in the <a href='#about' class='underline'>About section</a>."

},

{

"type": "about",

"title": "Methodology: Spiking",

"text": "<strong>Spiking</strong> is based on the Log Prevalence Ratio (LPR). It measures the <em>magnitude</em> of overuse and identifies 'spiking hallmark' words of AI."

},

{

"type": "finding",

"title": "The Syntactic Fingerprint",

"text": "A high proportion of overused terms are <strong>Function Words</strong> (e.g., connectors), suggesting that AI detection relies heavily on syntax rather than just topic. See <a href='https://arxiv.org/abs/2506.01407' class='underline' target='_blank'>related research</a>."

},

{

"type": "finding",

"title": "Model Idiolects",

"text": "Divergence patterns vary significantly across architectures (e.g., GPT vs. Claude), confirming that different models possess distinct <strong>idiolects</strong>. Read the analysis in <a href='https://www.scientificamerican.com/article/chatgpt-and-gemini-ai-have-uniquely-different-writing-styles/' class='underline' target='_blank'>Scientific American</a>."

},

{

"type": "tip",

"title": "Deep Dive",

"text": "Toggle the <strong>Table View</strong> to inspect the specific Part-of-Speech (POS) tags for every divergent word found in the dataset. OPM is occurrences per million (tokens)."

},

{

"type": "finding",

"title": "The RLHF Effect",

"text": "Why these words? Evidence suggests human raters during training (RLHF) often prefer this specific 'style', reinforcing the overuse of buzzwords. Listen on <a href='https://www.sciencefriday.com/segments/chatgpt-ai-word-choice/' class='underline' target='_blank'>Science Friday</a>."

}

]