Jakob Kather at ESMO: LLMs and AI agents in oncology, from single-task models to autonomy
jnkath · x · 2026-10-23
At ESMO 2026, Jakob Kather (Institute for Medical AI, University Hospital Heidelberg) presented "Generative AI, LLMs and AI agents in oncology."
Evolution: from 2000s classical ML and 2010s single-task deep learning, to 2020 self-supervised foundation models, 2023 LLMs, and 2025 AI agents — single-task to multi-task to autonomy.
Governance: the ESMO ELCAP guidance classifies clinical LLM systems into three types — Type 1 patient-facing (symptom queries, education), Type 2 HCP-facing (clinical decision support, multilingual communication), Type 3 background (data extraction, alerting). The EVALLM framework, in progress, will define evidence requirements for each.
Takeaway: clinical implementation depends not just on innovation but on data infrastructure and effective governance.
More from Research
- Cell Systems Voices piece examines future of genomic foundation models in biology — mo_lotfollahi · 2026-10-23
- reuF2F benchmark: 244 Lean-formalized open math problems, best LLM solves only 109 — wellecks · 2026-10-23
- Meta trains an 8B model to pick tools better than a 402B Llama-4-Maverick — dair_ai · 2026-10-23
- Entropy doesn't track what matters in chain-of-thought, BlackboxNLP paper finds — gsarti_ · 2026-10-23
- CT-Tox predicts human toxicity from rat data at 0.74 rank agreement vs 0.35 baseline — david_van_dijk · 2026-10-23
- is-lesion: open-source MRI lesion detection with calibrated System-One judgments on HF datasets — init0 · 2026-10-23