Field notes from Lithuania's medical AI conference: local models win on patient privacy
EmilPi · reddit · 2026-10-21
A first-hand report from Lithuania's conference on AI Technologies in Medicine. Key takeaway: medical researchers genuinely run local models, since strict patient-data privacy leaves no alternative. Highlights include AIML.lt's open-weights multilingual medical QA models (laya-multilingual-lt-med, Qwen3.5-4B-LT-med), an MCP server helping geneticists evaluate chromosomal rearrangement breakpoints in rare diseases, local-model bulk translation of medical datasets into Lithuanian, and classical ML work on chest X-ray diagnostic reliability. Models mentioned: Qwen3.5-4B, Qwen3.6-35B-A3B, Qwen3.5-122B-A10B, medgemma-27B — often outdated due to long academic cycles. Regulators focused on funding and patient-data/application safety.
More from Infra
- RunInfra launches $10/mo coding plan: fastest open models across 19 agents incl. Claude Code — ycombinator · 2026-10-21
- Kimi K3 hits 391 tokens/s per user on TPU v7 with open-source vLLM megakernels — woosuk_k · 2026-10-21
- TPN Bench launches quantization service: I-matrix and quant-aware distillation for any HF model — const_reborn · 2026-10-21
- Turso unveils MVCC implementation to lift SQLite's single-writer bottleneck — glcst · 2026-10-21
- Railway pauses all new builds amid major outage, users ask if vibe coding is to blame — AaronBergman18 · 2026-10-21
- a16z: Superhuman AI hardware design is the easy part — manufacturing is the hard one — a16z · 2026-10-21