A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis
Summary
A dermatologist evaluates GPT-6 Astra and Claude Fable 5.1 against a 10-year-old CNN fine-tuned on ISIC dermatology images for skin cancer diagnosis. The CNN achieves 62% accuracy on 200 biopsy-proven lesions, outperforming the frontier LLMs at 42%.