IdeaLens
Detectors of whose ideas a document contains (IdeaLens) and who wrote it (ProseLens), plus their data. Code: github.com/RishanthRajendhran/IdeaLens
Paper • 2610.06778 • Published • 2Note IdeaLens: Detecting AI Ideas in Long-form Writing
rishanthrajendhran/IdeaLens
Text Generation • 33B • Updated • 316 • 3Note The main idea-level detector: reads a document's outline and judges whose ideas it contains. Demo: ideadetector.ai
rishanthrajendhran/ProseLens
Text Generation • 33B • Updated • 406 • 2Note The prose-level counterpart: reads the document itself and judges who wrote the words.
rishanthrajendhran/IdeaLens-NoParaphrase
Text Generation • 33B • Updated • 172 • 1Note IdeaLens trained on outlines as extracted, without the paraphrasing step.
rishanthrajendhran/IdeaLens-Qwen3.5-9B
Text Classification • 8B • Updated • 16Note Idea-level detector on Qwen3.5-9B; reads the whole outline.
rishanthrajendhran/IdeaLens-Qwen3.5-9B-PerItem
Text Classification • 8B • Updated • 15Note Qwen3.5-9B detector that scores each outline item and pools the scores.
rishanthrajendhran/IdeaLens-ModernBERT-L
Text Classification • 0.4B • Updated • 32Note Lightweight idea-level detector on ModernBERT-large; reads the whole outline.
rishanthrajendhran/IdeaLens-ModernBERT-L-NoParaphrase
Text Classification • 0.4B • Updated • 19Note ModernBERT-large detector trained on outlines as extracted.
rishanthrajendhran/IdeaLens-ModernBERT-L-RolesOnly
Text Classification • 0.4B • Updated • 16Note ModernBERT-large detector that reads only the outline's sequence of roles.
rishanthrajendhran/IdeaLens-ModernBERT-L-PerItem
Text Classification • 0.4B • Updated • 15Note ModernBERT-large detector that scores each outline item and pools the scores.
rishanthrajendhran/ProseLens-ModernBERT-L
Text Classification • 0.4B • Updated • 56Note Prose-level detector on ModernBERT-large; reads the document itself.
rishanthrajendhran/IdeaLens-LogisticClassifier
UpdatedNote Logistic regression over outline embeddings; runs on a CPU.
rishanthrajendhran/IdeaLens-LogisticClassifier-PerItem
UpdatedNote Logistic regression over item embeddings, pooled per document; runs on a CPU.
rishanthrajendhran/WildOutlines
Viewer • Updated • 1M • 398 • 3Note The training corpus: role-labelled outlines of about 1M English web documents, from WildAI.
rishanthrajendhran/IdeaShift
Viewer • Updated • 3.5k • 202Note Documents generated from prompts carrying more and more of a seed document's plan.
rishanthrajendhran/IdeaShift-X
Viewer • Updated • 30.4k • 110Note Human web pages in 24 languages, each with an AI-ideas and a human-ideas AI version.
rishanthrajendhran/TwiceTold
Viewer • Updated • 100 • 95Note Short stories people wrote from the outlines of AI-generated stories, with the originals.