Research · Updated 25 Oct, 10:45 pm IST
10-year-old CNN outperforms GPT-6 Astra and Claude Fable 5.1 on melanoma test
Why it matters for readers: Demonstrates that newer does not always mean better and that simple, targeted models can still outperform the latest multimodal AIs on some tasks.
- A dermatologist compared two frontier LLMs (GPT-6 Astra, Claude Fable 5.1) to a fine-tuned Inception v3 CNN on skin lesion images.1
- The CNN, fine-tuned for about twenty minutes on public dermatology images, correctly diagnosed 125 of 200 biopsy-proven lesions (62%).1
- GPT-6 Astra and Claude Fable 5.1 each named the correct diagnosis for 84 and 85 of the same 200 lesions (about 42%).1
- The test used clinical photographs and dermoscopic images drawn from the MILK10k public dataset.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started