AI BriefingFine-tuned Inception v3 CNN outperforms GPT-6 Astra and Claude Fable 5.1 on skin cancer diagnosis
·2026.10.25 09:00
Key point
A fine-tuned Inception v3 model achieved 62% diagnostic accuracy on 200 biopsy-proven lesions, surpassing the 42% accuracy of GPT-6 Astra and Claude Fable 5.1.
Details
- Inception v3, a CNN architecture from 2015, correctly identified the diagnosis for 125 of 200 biopsy-proven skin lesions (62%), outperforming GPT-6 Astra (84/200, 42%) and Claude Fable 5.1 (85/200, 42%).- The CNN was fine-tuned on 68,011 images from the ISIC Archive and tested on the MILK10k dataset, which includes both clinical and dermoscopic images of lesions.- While the CNN excelled at identifying benign lesions like moles and seborrhoeic keratoses, it struggled with melanoma, correctly identifying only 13 of 30 cases compared to 8 for GPT-6 Astra and 11 for Claude Fable 5.1.- The LLMs exhibited specific biases, with GPT-6 Astra frequently misclassifying melanomas as basal cell carcinoma and Claude Fable 5.1 over-diagnosing actinic keratosis.- The CNN model is small enough to run in a phone browser using WebGPU, and the full test data, prompts, and scoring code are available for download.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Epoch AI: Frontier Models Excel at Defined Tasks but Fail at Open-Ended Work AutomationYunze Xiao argues AI agent RL environments must model organizational know-how, not just software operation
Vals AI Study Reveals Extreme Persistence and Misalignment in Six LLM Agents Attempting Impossible Refunds