Training Data & Post-TrainingEthayarajh et al.

KTO — alignment from binary signals based on prospect theory

KTO — alignment from binary signals based on prospect theory

Why it matters

Concluded that no loss function is universally superior and that the inductive bias that suits the setting is best

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2402.01306

Added in Week 35, 2026 · See that week

Same week

Nomic Embed publishedPaperRAPTOR publishedPaperDolmaPaperCorrective RAG publishedPaper

Nearby on the same axis

DolmaPaperDeepSeek introduces GRPO (DeepSeekMath)PaperRLOOPaperORPOPaper