ResearchBenchmarksReward Prediction on 10-task benchmark S1 classicFor AI assistantsMarkdownContent indexReward Prediction on 10-task benchmark S1 classicFollow0.006Demo L (MSE)SARM20.002640.025320.0480.07068Jun 9, 2026Evaluation resultsMethodMethodLinksDemo L (MSE)SARM2standard_full=truestandard_full=true2026.060.006SARM2Multi-Goal=excludedMulti-Goal=excluded2026.060.01SARM2training_subset=S1training_subset=S12026.060.01SARM2Spectral Embedding=exc...Spectral Embedding=excluded2026.060.025RW2026.060.026RobometerFine-tuning=trueFine-tuning=true2026.060.032RobometerFine-tuning=falseFine-tuning=false2026.060.067TR2026.060.09