The system Mythos Preview suggested a better next step than human researchers 64% of the time when evaluating research sessions that had gone off track; this is a significant improvement over 22% in 2024. The study also involved Claude, which is an important advance in measuring and practically applying the effectiveness of research-supporting AIs.
Mythos Preview often gave better suggestions for research steps than human researchers
The system Mythos Preview suggested a better next step than human researchers 64% of the time when evaluating research sessions that had gone off track; this is a significant improvement over 22% in 2024.
