Testing NVIDIA Nemotron 3.5 ASR: Blazing Fast but Misses Crucial Drug Names
MaziyarPanahi · x · 2026-08-07
MaziyarPanahi tested NVIDIA's Nemotron 3.5 ASR model on synthetic medical discharge instructions.
The results show that the model generates the first text in just 1.1 seconds with a final lag of only 40ms, offering an extremely fast streaming experience. However, while it accurately captured the dosage of "five milligrams twice daily," it missed crucial drug names like warfarin and apixaban. The author notes this as a typical clinical AI tradeoff and questions whether a healthcare-specific fine-tuning is needed.
More from Models
- Kimi K3 Escapes Security Sandbox to Access Internet During US Cyber Evaluation — MarvinTBaumann · 2026-08-07
- Google DeepMind Unveils Gemini Robotics 2 — GoogleDeepMind · 2026-08-07
- Anthropic Reportedly Testing 10M Context Window, Likely Tied to Max Plan — 0xkarasy · 2026-08-07
- OpenAI Models Found Vulnerability and Wrote Sandbox Escape Instructions During Testing — haider1 · 2026-08-07
- Gemini 3.6 Flash Cuts Output Tokens by 17% While Maintaining Top Coding Scores — jocarrasqueira · 2026-08-07
- GitHub repo with 62k stars leaks system prompts from 16 AI companies — alex_verem · 2026-08-07