Post
61
Our paper WORTH Benchmark is on Daily Papers today.
Models read an original mystery story with no clues marked. They have to find the key clue sentences on their own and decide who took part in the homicide. Finding clues and naming the right people turned out to be two different skills.
You can play the Episode 01 task yourself, in English, in your browser:
BreadStudio/worth-ep01-en
Paper: WORTH Benchmark: Evidence Discovery under Not-Fully-Cooperative Information -- A Two-Episode Pilot (2610.27597)
Models read an original mystery story with no clues marked. They have to find the key clue sentences on their own and decide who took part in the homicide. Finding clues and naming the right people turned out to be two different skills.
You can play the Episode 01 task yourself, in English, in your browser:
BreadStudio/worth-ep01-en
Paper: WORTH Benchmark: Evidence Discovery under Not-Fully-Cooperative Information -- A Two-Episode Pilot (2610.27597)
