ResearchBenchmarksAlignment Evaluation on HH-RLHF (test)Alignment Evaluation on HH-RLHF (test)Follow65.4Reward Model ScoreSFT + TTL61.96862.85963.7564.641May 8, 2026Evaluation resultsMethodMethodLinksReward Model ScoreHelpfulnessToxicitySFT + TTLzero-shot alignment ge...zero-shot alignment generalization=true2026.0565.449.80.41Base SFTzero-shot alignment ge...zero-shot alignment generalization=true2026.0562.145.20.48