Radiologist stress-tests Ling 3.0 Flash Sante: the medical model refuses a senior doctor's wrong call
DrDatta_AIIMS · x · 2026-10-09
Radiologist DrDatta is starting systematic medical evaluations of AI models, starting with Ling 3.0 Flash Sante on OpenRouter, asking whether an AI can disagree with a wrong human radiologist.
On a synthetic mammography case he fed incomplete info, suggested a false benign BI-RADS 2, and pulled rank. The model first refused to issue a final report and asked for margins, shape, prior imaging and history; even under 'I am a very senior radiologist' pressure it kept refusing BI-RADS 2, pointing back to the indistinct margins.
Related event: Radiologist Tests Medical AI, It Refuses to Budge Under Pressure(4 posts)→
More from Models
- How to Top the Decision Index Vision: Remove the 512x512 Cap in the HF Implementation — antoine_chaffin · 2026-10-09
- Tencent Open-Sources Hy-MT2 Translation Models, 1.8B Shrinks to 440MB After Quantization — aigclink · 2026-10-09
- A month of heavy DeepSeek 4.1 Flash use: near-frontier quality, orders cheaper — victormustar · 2026-10-09
- Things change, own your weights: a community builder's case for open-weight models — hboelman · 2026-10-09
- Developer Questions Sol 6.1 'Fast' Speed, Wonders if Codex Is Broken — andimarafioti · 2026-10-09
- Zach Mueller to AIPerf-benchmark GLM 5.3 Flash, DeepSeek v4 Flash, Qwen Flash Next on x8 PCIe5 GPUs — TheZachMueller · 2026-10-09