Training Data & Post-TrainingAhmadian et al.

RLOO

RLOO — a return to REINFORCE, arguing that much of PPO is unnecessary

Why it matters

Representative of the line of work that simplifies RL

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

Paperarxiv.orghttps://arxiv.org/abs/2402.14740

Added in Week 35, 2026 · See that week

Same week

Genie publishedPaperGemma 1 (2B/7B), under Gemma's own terms of usePressYang Zhilin frames AI as scaling laws and long context as key to AGIPress

Nearby on the same axis

DeepSeek introduces GRPO (DeepSeekMath)PaperORPOPaperKTO — alignment from binary signals based on prospect theoryPaperDolmaPaper