Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Preference Alignment on U10 (held-out evaluation set)

610.7Delta Reward

Robust PL

-24.428140.461305.35470.239Jul 2, 2026
Updated 22d ago

Evaluation Results

MethodLinks
2026.07
610.771.928.94841.363.224.3
2026.07
584.669.527.748.839.362.422.5
2026.07
48468.425.846.543.664.626.5
2026.07
482.167.22543.838.962.122.3
2026.07
46765.623.845.939.162.121.8
2026.07
426.765.224.246.140.162.422.6
2026.07
407.4662547.538.26323.8
2026.07
350.164.110.518.838.859.516.3
2026.07
341.361.79.418.637.15712.5
2026.07
33362.910.217.837.359.117.2
2026.07
261.561.75.914.836.358.515.2
2026.07
252.561.77.815.83556.812.1
2026.07
189.456.27.116.933.254.88.5
2026.07
0505.914.531.558.615.5
2026.07
05024.642.624.651.62.6