ResearchBenchmarksText Generation on Human Evaluation 6 tasks (10 pairs sampled per task)Text Generation on Human Evaluation 6 tasks (10 pairs sampled per task)Follow0.311Fluency Win RateADS0.20180.230150.25850.28685May 8, 2023Evaluation resultsMethodMethodLinksFluency Win RateFluency Loss RateFluency Tie RateFluency Zeta Score (ζ)Coherence Win RateCoherence Loss RateCoherence Tie RateCoherence Zeta Score (ζ)Relevance Win RateRelevance Loss RateRelevance Tie RateRelevance Zeta Score (ζ)ADSSteps=20Steps=202023.050.3110.150.5390.6660.3280.1720.50.740.2670.1560.5770.817ADSSteps=2000Steps=20002023.050.2060.1380.6560.8590.2170.1280.6550.710.2780.1610.5610.878