Report
AIM reportedly beats ScientistOne on two task groups
A post describing a Google paper says AIM ranks idea themes and audits whether the code matches each idea.
TLDR
A user says a Google paper describes AIM, a research agent that explores both strong and untested idea themes. An auditor rejects gamed results and relabels ideas when the code no longer matches them. The user reports that AIM beat ScientistOne by 1.6 and 4.9 points on two task groups and matched its best score up to 3.1x sooner.
Combined views
—
2 Sources, first seen ago
