Curated by Martin Weiss · 5 readings · Updated Oct 19, 2026
AI scientists
Start with the overall research workflow: a controller coordinates task drafting, experiments, verification, and reporting. Compare its orchestration choices with the systems below.
MIRA separates the decision about what to investigate from the execution of each investigation. Read for its treatment of research allocation and learning from partial progress.
The harness stores research state and gives each agent a fresh, role-specific context. Read alongside MIRA to compare how the systems preserve evidence and assign the next experiment.
Agents optimize solutions alongside inexpensive proxy evaluators, with fidelity checks against the true objective. This adds evaluation cost and proxy reliability to the comparison.
A conceptual proposal about allocating scarce scientific resources, assigning credit, and maintaining accountability. It broadens the systems papers into questions about institutions and incentives.