Reaction
A method is claimed to predict future SWE-bench rankings before post-training
A post asks whether the method could identify the best coding-agent checkpoint without post-training every base model.
TLDR
A post describes a method for spotting coding-agent potential before post-training. Its author claims it predicts future SWE-bench rankings with ρ=0.964, raising the question of whether every base model needs post-training to find the best coding-agent checkpoint.
Combined views
3.9K
2 Sources, first seen ago
