Show HN: DogCatTrace – Clapt was distilled from some random dog

Category: other

Tags: satire, model-fingerprinting, distillation, synthetic-data, web-app

Score: 5.0/10 (Innovation: 5, Technical: 6, Documentation: 7, Utility: 2)

DogCatTrace is a satirical tool that uses a contrived fingerprinting method (Hellinger similarity + softmax) to assign a fictional dog or cat breed to an LLM based on its integer outputs. It is a joke project demonstrating that distributional similarity does not imply model distillation, with a complete pipeline from synthetic data generation to a web interface.

Target audience: AI researchers, ML engineers, and anyone interested in model attribution or skeptical of fingerprinting claims

Repository: https://github.com/xiebaiyuan/DogCatTrace · HTML · MIT · 2 stars

View on Hacker News