Loading today’s stories
other
Third-party research cited as the source of the adapted figure showing RL improving out-of-distribution performance while SFT degrades under equal-compute post-training.
No evidence-backed relationships are recorded.
No current published clusters are mapped here yet.