Announcement
SynScience unveils FrontierML, testing AI research without a peek at the final score
SynScience says FrontierML’s 25 research problems require agents to devise and test their own progress measures.
TLDR
Bring your own yardstick: SynScience says FrontierML’s 25 open-ended machine-learning problems require agents to build and assess their own measures of progress. Submitted models or methods then face a held-out evaluation. @rohanpaul_ai reports that none of five tested frontier models reached the strong-result threshold on any problem.
Combined views
23K
5 Sources, first seen ago

Comments