254 points123 comments
Decade-old CNN beats top LLMs in skin cancer diagnosis
What is A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis?
A fine-tuned Inception v3 CNN from 2015 correctly identified 62% of 200 biopsy-proven skin lesions, outperforming GPT-6 Astra and Claude Fable 5.1 which only hit 42%. The test used paired clinical and dermoscopic images from the MILK10k dataset, and the CNN was fine-tuned in about twenty minutes.
Who it is for
Dermatologists and medical-AI developers should note that purpose-built vision models still surpass generalist LLMs for image-based diagnosis.
How to try it
Read the full analysis at https://lynchlab.net/llm-skin-cancer