ATHENA Masters 212 Biomedical Tools via Two-Stage Self-Learning
marinkazitnik · x · 2026-07-07
ATHENA is capable of identifying missing information and invoking the correct tools at each reasoning step, covering 212 biomedical tools to ultimately form an inspectable, complete reasoning chain. Because manual annotation at this scale is infeasible, training was divided into two stages: the first stage involved agents autonomously generating tool-calling trajectories for supervised fine-tuning; the second stage utilized reinforcement learning with scientific feedback rewards to further optimize the policy, enabling large-scale training without human annotation.
More from Research
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Artificial Analysis breaks down all sub-evaluations in Intelligence Index v4.3.2 — ArtificialAnlys · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27
- Meta FAIR's Self-Direction turns agent failures into reusable LoRA skills, +13.8pp on Terminal-Bench — karen_ullrich · 2026-10-27
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Why don't more LLMs ship 10M-token context? A researcher asks about attention tradeoffs — agarwl_ · 2026-10-27