"You're relatively right!" heatmap study of LLM fingerprints
Summary
Typebulb’s 'You're relatively right!' analyzes writing fingerprints across multiple language models using a word-based heatmap. It includes a large TypeScript/React bulb with gram counting, cross-entropy, and KL divergence computations, plus an embedded baked data block of model results and a data-rich discussion on evaluating model similarity across labs and models. The piece emphasizes reproducibility, tokenization considerations, and interpretive caveats when comparing AI writers.